home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Wildcat Files 2
/
The Wildcat Files 2 (Arsenal Computer).ISO
/
netmail
/
provider.faq
< prev
next >
Wrap
Internet Message Format
|
1994-12-10
|
209KB
Date: Thu, 17 Nov 1994 23:03:51 -0800
From: David H Dennis <david@amazing.cinenet.net>
Subject: The Internet Access Provider FAQ
THE INET-ACCESS FREQUENTLY ASKED QUESTIONS LIST
By David H Dennis (david@amazing.cinenet.net; backup e-mail dhd@smartdocs.com,
dhd@netcom.com).
[Third release edition: 17 November 1994]
Welcome to the third edition (and first formal release) of the Inet-
Access Frequently Asked Questions List (FAQ). Thanks for joining me
and my many contributers on this voyage into the unknown. Here,
we will see the traffic cops and maintenance people of the
Information Superhighway make their first struggling appearances.
Perhaps you want to be one of them, or maybe you're just curious.
Either way, this FAQ's for you.
A FAQ is very often prepared by someone who is not yet an expert, but
who wants to take the effort to become one, and this document is no
exception to the rule. I run a small provider called David's
Amazing Internet Services (DAIS), presently operating out of
my home on a 28.8kbps SLIP connection. As a result, I have not
encountered a Router, nor have I seen or paid for things like T1
lines or CSU/DSUs. Fortunately, I've had plenty of help from people
who have. This FAQ could not have been produced without the aid of
many people who have contributed their time, effort and observations,
either through private mail or the inet-access mailing list.
I'd greatly appreciate your contributions; please drop me a line at
david@amazing.cinenet.net. My provider is meant to be an innovative
combination of bulletin board and ISP; please feel free to give it
a ring at (818) 997-7500 or telnet to amazing.cinenet.net. Access
is free, so please feel welcome to come on board and stick around
for a while.
Sections marked with an asterix (*) are in special need of assistance
from some kindly soul with knowledge I just don't possess.
I would like to thank the following people, whose efforts were
particulary informative, helpful or encouraging:
Karl Denninger (Karl@MCS.COM) has been kind enough to share full
information about his equipment configuration with the list, as well
as contributing intriguing observations on the politics of the net.
I can't afford to duplicate his equipment, {sob}, and I certainly
don't know how to dig into the swamps of Internet politics like he
does, but reading about both is both interesting and useful.
Draper Kauffman <draperk@io.com> was kind enough to start an excellent
discussion of the pitfalls and problems of operating a small IP
provider. His discussion forms the core of my section on business
problems, which I'm sure will be of great interest to many of you.
Draper is the President of Illuminati Online, an Austin, TX provider.
Alicia Salomon <zone@access.digex.net> was kind enough to contribute
information on many different topics, including organizational and
administrative issues I have yet to become familiar with. She also
gave me considerable moral support and encouragement when this
project was just a gleam in its creator's eye. Alicia is an
independant consultant working on Internet-related issues. In
concluding one of her notes, she reminds us, "Also, something really
important I forgot was you should include a pointer to alt.caffine."
Consider it done, Alicia. :-)
My information on connecting to the Internet via leased lines came
mainly through the i.net info files (ftp to i.net:/pub/internetworks).
My thanks to Aaron Nabil of i.net for putting together the clearest
explanation I've seen of what equipment and services are needed to
start up a full Internet connection. Several books I've read have
not been as clear as his simple description of the requirements, as
shown in his info-prices file.
Most of the material in the BBS sections was kindly contributed by Mr
Eric S. Raymond <esr@snark.thyrsus.com>. Mr. Raymond is the
programmer and designer of Chester County InterLink (CCIL), a free
public access site in Chester County, Pennsylvania, USA. He is also
the updater and maintainer of the legendary jargon file, now
published as 'The New Hackers' Dictionary, Second Edition', a book I
bought and read with considerable enthusiasm. (The book brings back
all sorts of nostalgic memories of the good old days I spent hacking
as a "Tourist" at MIT with the much-beloved and now lamented
Incompatible Timesharing System (ITS).)
This FAQ is copyright (C) 1994 by David H Dennis. Unlimited non-
commercial redistribution is allowed so long as the text and copyright
notice is retained in its entirety and not modified in any way.
Commercial use is prohibited without written permission from the
author.
Release History and Changes
First Release: Sunday 14 August 1994 13:20:26
FAQ written. Circa 527 lines.
Second release: Sunday, 21 August 1994 12:52:10
Added distilled wisdom on low-capital IP startups; Added appendix on
use of BSDI Unix as a router; added statement on Net-99 courtesy of
Karl Denninger. FAQ balloons to over 2,000 lines for the first time.
Third release: Minor changes, unrecorded by history
Fourth Release: 17 November 1994
Added contributions on BSDI Unix by Tony Sanders and other
suggestions from him, particulary much information on how to use a
BSDI system as a router, and other generalized BSDI hardware
recommendations. Added information on credit card acceptance.
Rewrote the entire CIX section in view of recent channges to its
status. A little more on 'Guerilla Financing'. A section on
competition from the government, cable TV and telephone companies was
added. Many detail updates were also made that I don't remember.
Alas. :-( FAQ hits 4,119 lines, 197,862 bytes.
This FAQ was prepared using various computers and software,
including a Sun 3/60, an IBM ThinkPad 750C, Sager and AST laptop
computers and 486 DX2/66 computers running both Linux and OS/2. Both
GNU Emacs (Unix) and Epsilon (OS2-DOS) text editors were used. The
author most heartily endorses writing on the Sun (when at home) and
the ThinkPad (anywhere else); both are splendid machines for writing,
perhaps the best available.
TABLE OF CONTENTS
Chapter / Section Name Line Number
1 Introduction 00248
1.1 What is an Internet Service Provider (ISP)? 00250
1.2 Why might I want to become an ISP? 00320
1.3 Why might I not want to become an ISP? 00326
2 What about business organization and raising money? 00357
2.1 What sorts of business organizations should I use? 00359
2.2 What resources can I use to get capital for my ISP? 00408
3 What sorts of returns can I expect making as an ISP? 00488
3.1 Introduction: The world is changing. 00490
3.2 How do things pencil out? Some reasonably hard numbers 00509
3.3 What if I oversell my connection? 00619
3.4 Draper Kauffman's 56k Connection Sudden Death Scenerio 00718
3.5 So, what can we do about this? 00816
4 The Big-Time Competition: Should you worry? 00847
4.1 Invasion of the Federal Government 00849
4.2 Invasion of the Phone and Cable People 00882
4.3 What about public libraries and Freenets? 00965
4.4 IBM and Microsoft, the Terrible Two 01001
5 Equipment 01035
5.1 What kind of computer(s) do I need to become an ISP? 01037
5.2 Tony Sanders' Recommended Equipment List 01248
5.3 What equipment is needed to hook up my system to the Internet? 01492
5.4 What other equipment do I need? 01628
5.5 How many phone lines do I need? 01757
5.6 Where do I put all this stuff? 01828
5.7 This is so expensive! Where can I find this stuff cheap? (*) 01871
6 Hooking up to the Internet 01899
6.1 How do I hook up to the Internet? 01901
6.2 What are the levels of connection you can buy? 01907
6.3 What is Frame Relay? 01937
6.4 What are the main national providers, and how much? 01995
6.5 Where could I get a list of national and local providers? 02020
6.6 What about a SLIP connection? 02037
7 What are interchange agreements, and why should I care? 02052
7.1 What is the Commercial Internet Exchange (CIX)? 02054
7.2 What is NET-99? 02237
7.3 What is Metering, and why are people so emotional about it? 02268
8 Internet Software tips, tricks and answers 02323
8.1 The Permissions headache 02342
8.2 Which news software should I run? 02355
8.3 Let's back up for a minute. What IS news, anyway? 02375
8.4 Which mail transport agent should I run? 02393
8.5 How do I set up a Gopher server? 02461
8.6 What about the World Wide Web? 02489
8.7 What about Internet Relay Chat (IRC)? 02530
9 What about Fees, Terms and Conditions? 02594
9.1 How much can I charge? How much do other providers charge? 02600
9.2 How can I distinguish my service from the competition? 02633
9.3 Accounting and Billing Practices (*) 02643
9.4 Credit Cards (*) 02672
9.5 Making the Internet more User-Friendly 02794
9.6 What are SLIP and PPP? 02800
9.7 What about proprietary GUIs, such as the Pipeline? 02840
9.8 What about BBS Interfaces? 02852
9.9 Is there an easy, painless way to provide Internet services? 02926
10 What sorts of technical problems should I expect? 02974
10.1 What can be done about System Crashes? 02993
10.2 What can be done about Network Outages? 03018
10.3 Hung Modems 03083
10.4 What would be a good backup policy 03100
10.5 What services are particulary hard on performance? 03111
10.6 What can be done about users who walk away from the keyboard? 03167
10.7 What can be done about users who never log out? 03194
10.8 Trouble Ticket Systems (*) 03223
11 Who needs and wants Internet Services? How can we reach them? 03229
11.1 Graduating College Students 03234
11.2 People with Internet access at work 03243
11.3 Current callers to computer bulletin boards (BBSs) 03253
11.4 The General Public 03274
11.5 Some low-cost marketing ideas (*) 03286
12 Now that I have users, how should I deal with them? 03329
12.1 How should I introduce my users to the unique Internet culture? 03335
12.2 What sort of disclaimers do I need to protect myself? 03362
12.3 Can minors be allowed to use unrestricted accounts? 03366
12.4 How can I prevent minors from seeing the sex stuff? 03380
12.5 How should I deal with Netiquitte violations? 03403
12.6 How should I deal with disputes between users? 03439
13 Legal Issues 03449
14 Books and other Resources 03481
15 Glossary of Common Terms and Acronyms 03507
16 Resources 03653
16.1 Internet Mailing Lists 03655
16.2 Internet Newsgroups 03688
16.3 USENET Oriented News Groups 03721
16.4 Books 03742
17 Conclusion 03775
--- Start of FAQ ---
+ Introduction
+.# What is an Internet Service Provider (ISP)?
An Internet Service Provider (ISP) is a company that connects members
of the general public to the Internet. It can be distinguished
from an Information Service such as CompuServe or America Online by
its emphasis on Internet tools such as USENET News, Gopher, WWW, etc.
Traditional bulletin board systems (BBSs) normally don't have direct
access to the Internet and can offer only limited USENET news and
mail, with no other Internet services. I consider any online system
a provider that has a direct Internet connection and provides access
to it in one or more of the following ways:
* Unix Shell Account - Users are given the infamous "%" prompt or
some variation thereof. Sometimes a simple menu is also provided,
often cobbled up through various Unix utility programs. However,
the basis of the system is Unix, and normally the user cannot take
full advantage of the services offered without knowing at least a
few Unix commands.
* Provide SLIP, CSLIP or PPP access to customers. This lets them
connect up as an Internet host, using their own software. Macintosh
and Microsoft Windows devotees particulary like this approach.
This is also the only way to get pictures directly through the
World Wide Web, although it has some disadvantages as we will
see later.
* Provide customers with a custom BBS with specialized Internet
features (newsreaders, etc).
Many people have tried to put together some form of Internet access
under commercial or shareware DOS or Windows BBS software. Most that
I've seen have not been notably successful at this task. In
particular, the lack of high-quality newsreaders for USENET makes the
systems incredibly confusing to use. The advent of off-line mail
readers may help this at least somewhat; the majority of postings
through BBSs come from them. Unfortunately, most off-line readers
tend to mangle headers badly and are a major source of annoyance (and
occasional amusement) among USENET readers. The popular QWK format
is a particulary bad victim of this practice; it uppercases all
subject lines and limits them to 20 characters, both considered
extraordinary poor form by USENET readers. The Whaffle DOS and
Unix-based BBS has created a different off-line reading format which
I believe may work better for USENET applications; unfortunately, I
have not yet seen it in operation. Most Whaffle operators, however,
find that users consider their systems confusing and difficult to
navigate.
TBBS, a DOS-based BBS program, has announced Internet extensions this
August, although I don't know when they will be made available to the
public. Major BBS presently has an Internet module with severe
limitations; for example, mail and news processing must still be done via
UUCP (!). Based on the track records of the respective packages, I would
assume that TBBS' package will be the better one. However, I think the
best packages will probably be developed for Internet users by Internet
users, and that almost certainly means use of the Unix operating system
instead of existing commercial BBSs.
Incidentally, we have the usual dispute over what names to give ourselves.
Sean Shapira <sds@jazzie.com> and others would like to call us Internet
Access Providers (IAPs): "Personally, I call them Internet Access
Providers. They are only one class of Internet Service Provider. Others
focus on providing services to the already-connected Internet community.
GNN and EIT are examples of this." Since the name of the mailing
list is inet-access, he may have a point here. Why don't I change
the name of the FAQ, then? Because I'd just finished changing all
my ISVs (Internet Service Vendors) here to ISPs! :-)
Anyone else feel strongly on this issue? If I get enough response
on this issue, I'm willing to crank up M-X replace string again. :-)
+.# Why might I want to become an ISP?
Growth, money and the glamour of it all. The sleepless nights, the
18-hour days, the opportunity to exercise your mind and get creative
in the provision of imaginative services for your users.
+.# Why might I not want to become an ISP?
The capital needed can be daunting to we start-up folks. True, it's
nothing compared to setting up a CompuServe or America Online, but
for the individual, it's a lot of money. For the group seeking
venture capital, it's a lot of pain and paperwork.
And then, once your dream starts being real, there are always the
sleepless nights and the infamous battles with Sendmail and
InterNetNews. The sinking feeling you might get when you're staring
at a $ 2,200 router as part of a $ 3,000 a month connection, and
realize that you don't quite understand the beast.
The growth of SLIP/PPP accounts, which are very much a commodity
business, much like the phone or cable TV, may change the creativity
element beyond recognition for many providers. Still, the so far
incredible growth rates should help make up for this.
The whole question of who's really on the Internet and who's not
seems to be rather difficult to answer at this point. The Commercial
Internet Exchange (CIX) is filtering routes starting November 15th.
Or are they? This controversial and amazingly confusing issue is
discussed in a later section of the FAQ.
Finally, if your only reason to start a provider is to make money,
you probably shouldn't do it. You'll be beaten out by those of us
who love the net and who are willing to work utterly ridiculous
hours to make your system a success. If you don't love staring at
the screen for many hours, or if you value an offline social
life, the net's not for you.
+ What about business organization and raising money?
+.# What sorts of business organizations should I use?
For the budding ISP, the sole proprietorship is certainly the
simplest and most tempting organizational scheme. No tiresome
forms to fill out, no lawyers needed, and no thought wasted
on the issue.
Many people, however, note the virtues of a Subchapter S
corporation. This neat little format has the limited liability
protection of a corporation, combined with the more favourable
tax treatment of a sole proprietorship - you can still deduct
losses directly off your returns.
Unfortuantely, the liability shield, while it looks great on
paper, may not be quite so wonderful in practice. In the
case of corporations owned by individuals, the courts typically
"pierce the corporate veil", considering the individual and
the corporation identical in case of liability. So if your
solo venture goes bust, you may still wind up in trouble
despite all your fancy papers. Thus, to make Subchapter S
sensible, it's vital that you have at least one partner.
It's worth noting that any corporation, including Subchapter
S, requires a large commitment to maintaining curious rituals,
such as annual meetings and elections of directors. If you forget
even one small detail in these curious events, you may wind up
losing your corporate status. Beware!
Michael Krause <krause@en.com> summarizes the case for the
S Corporation:
* Tax benefits: "Give yourself no salary, take out loans from
business, etc. Works out perfect, a nice little loophole
in our tax system."
* Personal Asset Protection: You can take out a business loan
and leave personal assets protected - unless they are needed
to guarentee the loan. Unfortunately, most banks are too
smart for this little trick, unless your corporation has
substantial assets and an existance distinct from your own.
* <slightly> decreased liability. Note that corporations cannot
commit crimes; if your corporation is accused of criminal
behaviour, you as owner are responsible.
Like virtually all legal advice given on the net, this information
comes from non-lawyers. You should consult your own legal experts
before making any final decisions.
+.# What resources can I use to get capital for my ISP?
Obviously, the easiest way to find money is to use your own. No
tiresome investors to deal with, no people breathing down your
neck when profits don't come quite as fast as one might like, and
no delays waiting for people to make up their minds.
So, winning the lottery should be the ideal way to finance your
new business. Unfortunately, most of us will admit that this is
rather a long shot.
One solution to this problem is to have a full-time job already and
live quite frugally. Then the balance of your income can be used to
finance your system. I've managed to do quite a bit with this
method, but unless you're used to a frugal lifestyle or have a
princely income, it's really no fun at all. Worse, there are
definite limits to growth; I can afford to fund my 28.8kbps SLIP
forever; I couldn't afford even a month's worth of T1 charges without
outside help. Unfortunately, as we will find out a bit later in full
detail, this kind of plan may not be enough to start a profitable
provider.
Parents, friends, relatives and people who have seen you in action are
probably the best early investors. They know and trust you through
previous dealings with them. Creating a preliminary business plan
and walking through it with them will not only help persuade them;
it may also yield good advice about what is needed to sell it to
them or others.
A recent book called (if my memory serves) 'Guerilla Financing' tells
how you can locate "angels", individuals keen on investing small
amounts of capital ($10-100k) in start-up ventures. Their motivation
often comes as much from the glamour of being associated with some
prestigeous enterprise as a desire to generate profits; this is, for
example, the spirit in which Broadway shows are traditionally
financed. The book has some fascinating detail on the precise
procedures, and I will have to buy it next time I see it to give some
ideas and better pointers. In sum, however, an Internet provider
business might be a particulary good field to find an "angel", since
they are often interested in the glamour of investing in high-tech
ventures.
There are other, more formal, sources of investment.
Venture Capitalists. They take a sizable percentage of your business
for the money, but they might be your best chance. This is especially
true considering the "trendiness" of this subject. (Someone who has
actually dealt with a venture capitalist might be able to help me out
a bit with this section. :-) )
Note that venture capitalists are almost always interested in eventually
taking your company public and reaping the sizable returns that can
be obtained therefrom. If you don't want to eventually wind up with
the hassles of running a public company, you may not want to go this
route. Keep in mind, though, that it could make you a wealthy person
if everything goes right.
Banks. Not really an option until you're a going concern.
The US Small Business Administration (SBA). According to Joseph
Lamar Greer <jgreer@gladsonte.uoregon.edu>, it is nearly impossible
to get a SBA loan unless you can qualify as a minority. Many people
have gotten loans by taking on a qualified minority partner or
partners.
Lottery Funds and other Economic Development Grants: In some states,
grants are available from the state (often funded through lottery
proceeds) for business development. Note that this is in the form
of grants; you don't need to pay this money back.
Department of Commerce: The US Government is investing $ 1.2 billion a
year to promote "development and deployment" of the "National
Information Infrastructure (NII)". Whether this means anything that
can help us is open to question, but you can check out their FAQ
on this subject through their BBS (202) 482-1199, or their WWW page:
http://ntiaunix1.ntia.doc.gov:70/0/faq/niiques.asc
You would presumably have to be a non-profit or school to get these
grants.
+ What sorts of returns can I expect making as an ISP?
+.# Introduction: The world is changing.
For quite a while, you could start an ISP on $ 10k worth of equipment
and a smile. Karl Denninger's MCS-NET (mcs.com) did that and became a
sizable presence in Chicago very quickly.
Congratulate him. It's not so easy anymore, especially in areas with
present competition. You probably have your best shot if existing
ISPs have dreadful reputations. In fact, I decided to start my
provider after noticing Netcom's bad service; Netcom is the main
provider I know of for Southern California. (I am based in Van Nuys,
California, which is in the San Fernando Valley. I believe myself to
be the only full-access provider native to that area, although there
are a couple of BBSs that offer various forms of access to
newsgroups).
My thanks to Draper Kauffman <draperk@.io.com> for providing the
initial inspiration for this section.
+.# How do things pencil out? Some reasonably hard numbers
[NOTE: This is a brand new section of the FAQ, written by yours
truly with minimal input from the outside. I'd greatly appreciate
comments or corrections].
One of the most controversial aspects of Internet provider lore is
how much load you can put on your system and its Internet connection
before things become intolerable to your customers.
To run a high-quality, conservative service, the consensus seems
to be the following:
28.8K SLIP: You can run three phone lines and get reasonable, but
not great, service for all of them. This is what I'm doing at
present, and it does work well, although the lag in telnet connections
(caused by high latancy on the phone lines) is bothersome to many.
CSLIP (compressed SLIP) seems to produce a dramatic improvement over SLIP.
56k: You can run up to eight phone lines and get satisfactory service.
T1: Now we're motoring! A T1 line can support up to 27 times more
lines than a 56k - thus, around 200 phone lines.
Here are some rough rate calculations, together with the provider
that supplied them:
Provider Service Monthly Cost Line Cost Total Cost Cost/Incoming
------------- --------- ------------ --------- ---------- -------------
Cinenet 28.8 SLIP US$ 125.00 20.00 145.00 48.00
Internetworks 56k US$ 350.00 100.00 450.00 56.00
Net-99 T1 US$ 1,999.00 500.00* 2,499.00 12.00
* This number is off the top of my head; quotes I've seen for T1 connections
range from a shade under $ 400/month to well over $ 1,000.00.
Cinenet is my present provider for SLIP; Internetworks is a provider I've
been considering for 56k, and Net-99 is the provider I would most likely
use for T1 if I got that far. (Internetworks does not offer T1 in my
area; Net-99's 56k charge is a rather stiff $ 999/month).
The "Cost/Incoming" column is the cost of the connection per each
incoming dialup line. Each line should be able to accomodate about 10
subscribers, using the most conservative calculations.
It's easy to say that the T1 is by far the most appealing option here,
unless you don't have investors able to pay $ 2,500/month until you're
properly settled.
Let's dream a little and assume we could run each one of these options
at full capacity. The numbers run like this:
Connection Phone Rate/L Cost/ Cost/ Gross
Connection Cost/Line Lines Commercial Line User Income/L
---------- --------- ----- ------------ ----- ----- -------
28.8 SLIP 48.00 003 30 (bus) 78.00 7.80 100.00
56k 56.00 008 30 (bus) 86.00 8.60 100.00
T1 12.00 200 30 (bus) 42.00 4.20 150.00
/-------------------------------------------------------------------\
| A note to our European and Asian friends: Rates for you are much |
| higher. In fact, they're over TEN TIMES more than this chart! |
| I believe Net-99 is planning international expansion; you might |
| want to contact them if you're interested. |
\-------------------------------------------------------------------/
The cost per user assumes 10 users per line, the standard. Gross income
per line assumes rates of $ 10/month for 56k or below, going up to
15/month for T1 service. Running the numbers shows extremely low gross
profit for 56k or 28.8 SLIP connections using the standard measures.
28.8K SLIP deserves mention as a viable option for the start-up provider,
just to get its personnel familiar with the various pitfalls of running
a system connected to the Internet. I've learned a great deal with my
28.8, and recommend it very highly to anyone considering a start in this
business. As you can see by the table above, it's no moneymaker, but
it's not a dramatic money loser, either. Anyone with a more or less
decent daytime job could maintain a 28.8 SLIP provider for as long as
their interest continued.
The 56k option looks like an extraordinarily unhappy compromise. If
we subtract basic expenses of $ 86/line from the gross income of
$ 100/line, we get only $ 14/line in gross profit, or $ 112/month for
an 8-line system. Obviously, nobody's going anywhere on a 56k
connection unless they either charge a lot more than the going rate
as I know it, or bend the rules to the breaking point.
Finally, if you are lucky enough to be able to afford a T1 connection
to the Internet, and have a successful marketing plan, the T1 option
is very clearly an excellent viable business. If you could fill up
all 200 lines a T1 is capable of serving at a rate of $ 15/month,
you'd have 2,000 users and $ 30,000/month to dispose of. You'd probably
need to hire a few employees at that level - it couldn't be done
properly with just one person - but you'd be able to afford them.
Tony Sanders <sanders@bsdi.com> suggests fractional T1 as a good
way to 'ease in' to the real thing. "You pay full T1
line charges but the total cost is quite a bit less and it makes for
a very easy upgrade path." This seems to depend a great deal on both
the provider and the telephone company you have to deal with. When
I was talking to ISI Network Associates and Pacific Bell, I found that
ISI charged the same for Frac-T1 and T1, and Pacific Bell only charged
about 20% less for the smallest increment of frac-T1 (256k).
T1 people can also sell SLIP connections reliably, which are generally
significantly more expensive (in the $ 30/month and up category). SLIP
is generally a high-bandwidth eating operation, so normally you won't
want to sell this form of connection on a 56k or lower line.
+.# What if I oversell my connection?
Probably the most tempting option for the 28.8K or 56k provider who'd
like to get some decent profits out of his system is to oversell the
connection - that is, to exceed the recommendations listed in the
previous section.
The argument is seductive: Many people are cheap. They'd rather have
an inexpensive connection than one that worked perfectly. I (the
start-up provider) am just one person, and I can't provide a perfect
system in any case; I just don't have the capital for a T1 or a
24-hour staff. Could I play the ISP game anyway, by just selling an
overstressed connection for less money than other providers?
The main problem with this game is that it's too easy to play. If
you offer cheap service at cheap prices, there are bound to be people
with more resources than you who can offer cheaper service at even
cheaper rates. This is Karl Denninger of MCS.COM's comment: There
will always be someone who can undercut you. Quality service,
though, is very hard to come by in the Internet provider business.
Many very large providers, such as Netcom, are highly vunerable to
complaints about terrible service. If you can capture some of their
customers by offering excellent service - even for more money - you
probably have a very good shot at gaining market share, even over
very large companies. So you may want to at least consider the high
road, not the low.
Michael Krause <krause@en.com> is an excellent example of a provider
who's managed to succeed despite the handicap of a slow link to the
Internet. His system runs 8 phone lines on a 28.8K link. What makes
this possible is that only a few services - most notably FTP and
downloading images through WWW (which shell or BBS account people
can't do anyway) require a major proportion of the data pipe.
Mudders, Telnet and IRC users are low users of the system capacity.
Usually he sees a maximum of two simutaneous FTP sessions, which can
be handled by the system. He points out that, as long as his
customers don't have experience with faster providers, the speed is
not a tremendous issue with them. He suggests that the peak
number of users on a 56k would probably be between 75 and 100. A
couple of other people have mentioned 40 as a "reasonable" maximum
number of simutaneous connections on a 56k with light FTP traffic.
Note that heavy use of FTP and WWW may change this dramatically in
time.
>From personal experience, I know that users will stick around after
just about any disaster, as long as access is free, and it's
understood beforehand that the system is experimental. I knew that
my system would be unreliable due to the new software I was writing,
and my inexperience as a system administrator. As a result, I
started by running it at no charge, with the understanding that lost
mail, connection problems and such would be accepted as typical
experimental system pains. So far, the overwhelming majority of my
users are very loyal, but that may change when I start charging even
a nominal amount of money. In short, giving service with problems at
a low price may be a reasonable strategy, but unfortunately there's a
big difference between "low" and "free" in most people's minds.
A couple of my users have told me that they don't want to pay for the
system as long as telnet connections are so slow, so even my strategy
may not pay off. This may, however, be due in part to problems with
runaway processes I've had on my system, which have decreased
performance for all users. Since then, however, I've solved my
runaway problems and gotten a CSLIP connection, which is much faster,
and I seem to have silenced the doubters.
News and mail reading and writing, of course, use virtually no
bandwidth at all, and you could run a news/mail only system on a very
slow SLIP connection. However, the value of this to the general
public is questionable. Still, if this is the primary interest of
your users, they'll probably be pleased with virtually any bandwidth
level. Note, however, that even a 28.8k SLIP doesn't seem to be
enough for a full newsfeed through INN, at least through my present
provider. (Things have improved dramatically since I started using
CSLIP with them, but how much I'm not sure at this time).
Many people get Internet access through their work or school, but
those institutions normally censor the content provided. For example,
it's pretty tough to find alt.sex.bondage on a corporate machine,
unless you're its manager. This opens up a surprisingly large market
of people who have access already, but want to telnet in to check
out the "forbidden" topics and sites out of their school or employer's
wary eye. This is likely to grow with the recent decision at a major
university (I think it was CMU) to halt access to sexual newsgroups.
This might make a system with an excellent news connection and nothing
else a viable site. It's also an excellent way of competing with the
Freenets, which are cheap but heavily censored.
I have noticed that, as long as FTP works, the slow speed doesn't
bother me much. What does bother me is delays in character echo when
I type. This seems to indicate that, if your system is fast enough,
you may be able to deliver service that's perceptively better than
Netcom's even with quite a sluggish link.
My conclusion to all this is that many ISPs can get away with bending
the rules for a while, at least until the competition gets a T1.
Then, all that careful business planning and development may go to
naught. As Tony Sanders <sanders@bsdi.com> summarizes the situation,
"It's kind of like a race for the T1 connection :-)"
+.# Draper Kauffman's 56k Connection Sudden Death Scenerio
This message is both so scary and so dramatic that I can't resist keeping
it in the FAQ, even though I've summarized many of the financial arguments
above. The message is that if you have a 56k connection, you have at
most a $ 500 gross profit, even with a $ 20/month rate. And $ 20/month
is unsustainable in today's market. Here's what that $ 500 would cost you,
if you could even get it at all:
>From Draper Kauffman <draperk@io.com>:
How hard do you have to work to get that? To begin with, if you are
growing fast you've recruited a lot of new users recently. They'll have a
million questions and requests. You're also doing all the accounting,
putting out accurate monthly invoices for 100 erratic users, depositing
70-80 checks, and deciding what to do with the no pays and late pays.
Meanwhile, you're trying keep your system up and your 8 bargain modems
working, and deciding what to do when mail runs 2 days late, or your
newsfeed stops coming, or whichever of the normal Net-crises hits you that
day.
All told, you probably put in 50-70 hours per week, maybe a lot more. You
make less than a ditchdigger, and you aren't getting paid for the use of
your equipment, so you haven't got any source of cash for upgrading or
expanding your system.
But you're still adding customers! And here comes the crunch that
threatens almost every low-capital ISP these days: too many users asking
too many questions, not enough bandwidth, people bitching about how the
system slows to a crawl every time you get your newsfeed, programs crashing
because the user disk is full, and so on.
Too many problems are allowed to fester and turn into flames, and suddenly
you just can't cope. There aren't enough hours in the day or dollars in
the bank.
Angry customers quit and bad-mouth the system and new people stop coming.
If you keep your rates at a moderate level you won't have enough volume to
cover your costs. If you drop your rates to bargain levels, you keep more
customers, but you're losing money on every one, so that's no help. You're
bleeding cash, and pretty soon some unforeseen expense will put you out of
business. You probably won't even know it until your check to the telco
bounces, since you haven't had time to do the bookkeeping for months.
That's the nightmare. Even if you can raise more capital at that point and
try again, you have to fight the bad rep of having run a shoddy operation.
And bad word of mouth hurts you more on the Net than in almost any other
business.
What causes this scenario? Here are ten good reasons:
1. Having insufficient capital.
2. Underestimating the time and routine expenses involved in a startup.
3. Overestimating the owner's knowledge, abilities, and stamina--the
heroic programmer complex.
4. Grossly underestimating the manhours needed to get the system up and
keep it running smoothly under load.
5. Starting with inferior services: slow connection, inadequate disk
space, skimpy software (gopher, lynx, etc.), slow or partial
newsfeed, and/or running too many functions (news/mail/users/etc.)
on one CPU and drive.
6. Charging too low a price (as a result of points 2 through 5).
7. Using cheap hardware and no backups.
8. Starting with a system with insufficient capacity to produce enough
profit to finance continued growth.
9. Allowing growth to exceed the sustainable system capacity.
10. Sloppy and inadequate accounting--it's easy to get behind, and
usually fatal.
(I would add inadequate marketing to the list, except that that can easily
be a blessing. One highly successful marketing effort could add 2-300
users in a week, completely overloading the system.)
Here's Draper's summary of the present environment:
Although there is market to market variation, today's ISP startup faces a
significantly more difficult challenge than those that started in the last
few years. Quality expectations are up and user fees are down. The result
is that low budget/low volume/low quality/low cost systems can no longer
expect to find a profitable initial niche in most markets. Without new
capital or a steady stream of profits, they have no way to increase quality
or capacity
Increasingly, a new service in a competitive area needs to offer a full
range of services, good user support, and a fast, reliable, and accessible
system in order to charge a premium price. Success will require larger
amounts of capital or inventive ways to overcome the numerous barriers to
low-volume profitability.
+.# So, what can we do about this?
These are my personal suggestions as FAQ maintainer on what to do if you're
a low-budget IP startup and somewhat intimidated by all this:
(1) Create some innovative services. For example, I'm coming up with a
unique program to make it easier to access the net. I'm also developing
a fresh new Pipeline-like GUI system. A more viable example for less
technically oriented people would be to spend a few hours a week net
surfing and reading net oriented publications like Wired. Then, you
can announce the "newsgroup of the week", "URL of the week" and "Telnet
Site of the Week". It would also not be a bad idea to put out a monthly
newsletter that contained that information and told people to be sure
to come online for that and similar events.
I think if you make your system a valuable resource to show people what
they can do on the net, you'll build loyalty that will pay off when
the crunch comes. And you don't have to do that just through direct
personal communications; inexpensive media such as a simple informational
menu and newsletters will do fine.
(2) Try to find a backup source of capital. From what we've seen in
previous sections, it's highly doubtful that 56k is a viable solution,
especially if you need to share your returns with investors. As a
result, it's pretty much T1 or nothing. Costs for this seem to vary
dramatically depending on your market, as I've covered previously.
(3) Don't quit your day job. A 28.8k system will work fine for you
to test your ideas until you can get financing to get closer to the
big time.
+ The Big-Time Competition: Should you worry?
+.# Invasion of the Federal Government
The Feds helped make the Internet the incredible success it is
today, so it seems sensible at first blush to suggest that this
should continue. As you probably know, the government is in fact
abandoning the Information Superhighway, heading for the exits just
as the place is starting to look pretty darn impressive. Many people
have in fact wondered out loud why the government doesn't just buy
a bigger backbone and stay in the business.
The Internet was a highly successful government program mainly
because they didn't do much. Yes, they provided the funding, but the
key to the Internet's success was that they didn't put very tight
control over what it was used for. Thus the curious fact that a
sizable percentage of network traffic is alt.binaries.pictures.
erotica.female instead something sensible like scientific reports.
The Internet has been operated not by the government, but by local
sites.
Unfortunately, this non-governmental control combined by Federal
funding just cannot last. Now that the "Information Superhighway" is
becoming a more prominent part of people's lives, it's only a matter
of time before Fundementalist Christian groups try and get
alt.sex.stories booted off the net. And if our woozily incompetent
government is still in control, there's a pretty darn good chance of
that happening.
Best of all, the lack of government control over the net gives us the
ability to start our systems, run them as well as we can, and even
have a little fun doing it. Don't ask for the government to return,
or we might get the Post Office of Internet providers - slow,
expensive and stupid.
+.# Invasion of the Phone and Cable People
Cable TV companies represent a more interesting competitive threat.
For just $ 75 or so a month, they tell us, you can have a circa 56k
connection to the Internet, providing you with far higher quality
then you'd ever get through a traditional ISP. PSI recently
conducted a joint experiment with a cable company in Cambridge, MA.
Obviously this is a very biased location because more computer and
Internet lovers live there than virtually any other place; you would
expect firm demand here even if the whole thing was a disaster
everywhere else. (True, many people would have access through their
places of employment. Still, there are more people who would find
the sheer technological "win" of interest than anywhere else I can
think of). I have read that the venture was a disaster, with hardly
any signups. My suspicion is that few people wanted to pay those
prices, when they could get a $ 20/month account from an ISP.
However, it's also possible that people are using their free
university and work-based Internet accounts instead of going with
PSI. Many Internet users, however, normally use separate accounts
for work and home, so my guess is that the PSI service was just too
expensive for the benefits offered.
Before ending our discussion of cable TV companies, it's worth noting
the results of another venture, "video on demand". This was supposed
to be the holy grail of cable TV, the service that would pay all the
fantastic costs associated with the "information superhighway". A
distinctly low-tech test was created, where a rack of VCRs was hooked
up and a person hired to grab the requested tapes and load them.
Prices ranged from $ 0.99 to $ 4.00 per view. For a typical video on
demand venture to be solidly profitable, executives predicted that
roughly four rentals per month would have to be made per customer.
The actual figure was slightly under two, attained with the kind of
promotion blitz that couldn't possibly be duplicated on a wide scale.
Apparently people actually enjoy making the trip to the video store
to check out their evening's entertainment. (This was reported in a
recent issue of Wired (I think September 1994) and in the LA Weekly).
My conclusion from these two points of information is that the telephone
and cable-driven superstructure is likely to be a flop. Internet
services would require far more complex connections than video on
demand or home shopping, and I doubt that the profit potential is
as high. As a result, I doubt that we have much to fear from the
"cabledroids".
The phone companies may be more of a threat, but my suspicion is that
their bloated overhead levels will require high pricing and thus plenty
of room for Internet providers.
There is, however, one thing to watch out for: Many telephone
companies are arguing to drop the traditional local calling areas and
replace it with metered service. If they did that, and offered an
unmetered Internet connection, they could own the market. Be sure to
do your darndest to make sure this doesn't happen in your state!
ISDN might be another version of this, which should in theory be able
to offer switched 56k connections to the masses. So far, the phone
companies have been very sluggish to promote this service, but with
the ever-increasing demand for bandwidth this is bound to change in
time. Unfortunately, such an event is bound to hurt the typical ISP
big-time, since the T1 lines we buy (forget about a 56k when your
customers use ISDN!) will carry about 1/3 the users they did
previously. Whether the public is willing to pay the increased costs
associated with such a service is open to question, considering the
failure of early cable TV efforts discussed above.
ISDN has the potential to be a sneaky way for the phone companies to
get back into the measured service system they know and love. It
looks like they have lost the battle for measured service for voice
phone connections, but ISDN is a brand new game. Right now, Pacific
Bell makes residential ISDN available on a measured basis during the
day, and a non-measured one on evenings and weekends. We'll see if
people take the bait. (Another data point: Pacific Bell has recently
announced that its outrageous local/toll rates are going to be
decreased by 40%. It looks like the threat of competitive pressures
is working to reduce Pac Bell charges substantially).
Karl Denninger <karl@mcs.com> offers ISDN on his provider. When I asked
him how well it was working, and how good customer acceptance was, he
said, "It's dogs---; interoperabilty problems galore."
In short: Keep your eyes open, but at this point I don't think the
competition is likely to be as bad as it looks.
+.# What about public libraries and Freenets?
Well, this is a strange one. Here in California, we hear about
libraries shutting down for lack of support every day, and yet people
are talking about hooking them up to the Internet! I guess Pac Bell
will be donating the service or something, since otherwise that would
look like a pipe dream.
It's not really clear what the libraries will use their link for, of
course. Perhaps someone familiar with this issue could let me know.
There are two schools of thought among ISPs when it comes to library
connections to the Internet:
(1) They are evil, government-subsidized organizations who might
become ISP competitors "through the back door", giving free or very
inexpensive services.
(2) Why not join them, not beat them? Most libraries really don't
have the resources to run themselves, much less operate an ISP. Give
them space on your ISP, and let people have limited free access from
library-based terminals. Then, if they want more, or if they want
modem-based access, they will naturally come to you first.
However much I may dislike government-based organizations on the
main, I still think (2) is by far the best thing for an ISP with a
potential library competitor to do.
Freenets generally offer very limited Internet access. A large
number of my system's customers have come from the Los Angeles
Free-Net; they don't care for its censorship policies (which I
gather exclude the sexual stuff) nor for its very limited range of
systems you can telnet to. As a result, I think you could consider
your local Freenet a nice way to introduce people to the Internet
who you can then talk into becoming your customers.
+.# IBM and Microsoft, the Terrible Two
A convenient SLIP package is included in every copy of OS/2 Warp, the
latest incarnation of IBM's OS/2 operating system. Early returns say
that the Internet services part is a truly superb package, more than
competitive with other available Internet software. Information on
the Internet's OS/2 forums are showing IBM's software capturing the
hearts and minds of those who buy it.
At first blush, this looks like excellent news for Internet
providers, especially if Windows95 follows this lead. (Windows95 is
presently rumored to have Internet services, but they are said to be
far inferior to OS/2 Warp's).
Alas for us, IBM is going to compete as a provider, offering a full
Internet connection at what looks like a very reasonable price. If I
recall correctly, it's $ 12.95 for six hours or $ 30 for 30 hours.
This is ominously close to what normal ISPs charge. If IBM's service
is of high quality, it could make a serious dent in our industry,
especially with the members who had not converted from Unix shell to
SLIP/PPP accounts.
Microsoft is also getting into the act, with their upcoming "Marvel"
online service. They're going to charge $ 4.95/month plus some unspecified
hourly rate. This one doesn't worry me as much as IBM; it's meant to
be a new competitor for CompuServe and America Online, and will be priced
at hourly rates that are not generally competitive with ISPs. I think
the main purpose of Marvel is to try and wean their customers off of
phone support and on to cheaper (and potentially revenue-producing)
on-line methods. I would doubt that they are likely to offer full-
blown Internet access, at least not in ways that are competitive with
ISPs.
+ Equipment
+.# What kind of computer(s) do I need to become an ISP?
I don't know a frightful lot about hardware, but this section of
the FAQ has stood the test of time and comments pretty well. The
main exception is a great deal more information from BSDI fans
about their favourite system. I haven't heard at all from Sun
fans, but the occasional Linuxer has made his presence known.
If you have some comments on this information, please speak up;
I'd like to be able to flesh this section out a bit more.
To start with, you need some computer capable of running Unix.
Opinions vary dramatically over what operating environment is best,
but I don't know of anyone who uses anything but a Unix derivative
operating system. Here are a few capsule arguments for various
versions of Unix; corrections are welcome. Remember, the most
ferocious holy wars are often between those of strikingly similar
beliefs! Christianity versus Judiaism, Sun versus BSDI versus Linux.
Watch the flames grow!
Sun Workstations and SunOS. If you have a ton of money to blow, go
right over to your local Sun distributer and pick up a couple of nice
little SparcStation 20s. They're just a shade pricey, but net wisdom
has it that you'll have the least trouble if you go this route. Sun
owners are generally quite happy with their technology, but many
admit that the PC route is so much cheaper that the entire world is
going in that direction.
PCs running BSDI Unix, a BSD derivative. BSDI users are ferociously
loyal to their system. Karl Denninger, probably the most successful
provider on the Inet-Access mailing list, uses modified BSDI systems
with, if my memory serves, 64MB of RAM and a 1.0GB hard disk on each.
(Karl has not corrected my memory, and I'm sure he's seen a copy or
two of this FAQ). Each one can service approximately 64 users when a
terminal server is used. He has told the world that the system is
very solid and technical support is superb - a rarity among operating
systems, or any other software for that matter. Source code license
is $ 995.00; binary is $ 595. Once you buy either initial license, a
license for each additional machine costs $ 200. I believe they are
willing to negotiate a site license for very large numbers of
machines. Karl Denninger will probably tell you you need the source;
he's modified it extensively. In addition, Eric Raymond
<esr@locke.ccil.org> said, "Pay the $ 995. it's worth every penny",
and Mark E Mallett <mem@mv.mv.com> says, "I agree with the support
for the system." You can find out more about BSDI by mailing to
info@bsdi.com. The ability to use a BSDI system as a router may make
BSDI the system of choice for many providers; this could save you
about $ 1,300 or more, depending on configuration. So you could buy
a BSDI license for the money saved from the router alone.
PCs running Linux. Linux seems to be the number one choice among
providers coming up from the BBS world. It could be thought of as
the latest and best continuation of the "Hacker Ethic", the belief
that software should be free, and people should get the source and
play around with it. (For information on the "Hacker Ethic", see
Eric Raymond's <esr@snark.thyrsus.com> 'The New Hacker's Dictionary'
[Second Edition]. It makes me nostalgic for the years of my
adolescence, as an ITS PDP-10 hacker at MIT. See the appropriate
entries for details). The lack of $1,000 licensing fees for the
source code probably has something to do with Linux' success over
BSDI, as does the notorious cheapness of computer hackers. Support,
surprisingly enough, is excellent. Post a question on the newsgroup,
and you'll get friendly answers with good information within hours.
Although the networking code is infamous for problems, I've had
surprisingly little trouble with it. My present system networks a
Sun 3/60 with my Linux PC; the Linux PC is connected to the Internet
through a 28.8kbps SLIP connection. Although it wasn't frightfully
easy to connect, everything is now working surprisingly well, with
little trouble. (It's now been about 2 1/2 months since then; the
system has continued running very well). Linux has been used with
56k connections as well, and early reports appear to be positive.
The Slackware distribution of Linux is the standard and is highly
recommended. The best deal is probably the Trans-Ameritech CD for
$ 30; mail roman@trans-ameritech.com for additional information.
PCs running BSD derivatives other than BSDI. I know little about
this, so someone else will have to fill in this part of the FAQ. BSD
is said have better-debugged networking code than Linux, but support
from the newsgroup doesn't seem to be as good. I would say that Linux
has at least 10 times as many newsgroup readers, with proportionately
better support chances.
PCs running SCO (Santa Cruz Operation) Unix. This Unix exists, and
was basically the microcomputer standard for some time. It was
originally based on Microsoft's Xenix product, although I think it's
now a variant of Unix System V. It's quite pricey, but few who've
tried it would pick it over Linux. Even SCO's support, which you might
consider a major advantage of a commercial product, is rated as poor
by those who've used it. The bottom line is that either BSDI or
Linux would be better choices for a microcomputer Unix system.
Other Systems: The IBM RS/6000 and HP workstation users have a small
but vocal group of fans. However, again, I know little about them.
PCs running Windows NT. The overwhelming majority of Internet
administrators do not believe NT is the operating system of choice
for an Internet service provider. This is probably a rather kind
description compared to what you'd actually hear from a Unix user
confronted with a question about NT. It is a matter of record that
Microsoft uses Windows NT as their FTP server, named (with more hope
than sense, perhaps) "gowinnt". It is also a matter of record that
Microsoft.com, their mail server, is a SCO Unix box. Admittedly,
Microsoft Windows NT is not, perhaps, quite as unreliable as your
copy of Microsoft Windows. Hopefully. In any case, if they can't do
electronic mail through NT, when they have the strongest possible
reason to do it, you probably can't, either. Beware.
When you get past a certain number of users, you will need to network
several of these together. This is, for example, the approach taken
by Netcom, which now has 17 SparcStations, massively equipped. (Note:
Netcom's incredibly poor reputation for running sluggish systems seems
to have been caused largely by their lack of investment. The latest
upgrade to 17 systems from 14 helped quite a bit for a while, but
they are now once again heavily bogged down).
To start out in a very small way, you need at least 1 GB of hard disk
space. To store USENET news for any appreciable amount of time,
you'll need many times that. I can store about three weeks' worth of
ALT.* on the one 1.8GB Quantum hard disk I have dedicated to alt news.
(This includes the binaries groups, however, which you might want to
expire more quickly or not carry at all).
Jim Dixon (jdd@aiki.demon.co.uk) strongly recommends SCSI disk drives
and controllers; he is, of course, correct. You will wind up being
virtually forced into this route anyway, since most large hard disks
are available only as SCSI devices.
Eric Raymond <esr@locke.ccil.org> adds: "We started out with IDE disks
to save money. *Big* mistake; we had nothing but grief from the cheap
IDE controllers we were using. Go SCSI from the beginning; the
reliability and performance is more than worth the slight premium."
Sean Shapira <sds@jazzie.com> has another viewpoint. "Linux
will support 4 IDE drives, and these are commonly available at 500MB,
for a 2GB system. Beyond that, a wise provider might consider distributing
the load to a second server anyway. Admittedly IDE is slower than SCSI,
but I argue the trade off may be worthwhile." (I wrote back to him
noting that there are now 1.0GB IDE drives; he doesn't recommend them
because they are not yet as cost-effective as the smaller units).
Jonathan Heiliger <loco@netsys.com> notes that these are EIDE
(Extended IDE) instead of IDE drives; BSDI doesn't presently support
them. So if you're a BSDI fan, you'll have to wait on this one. (He
is, of course, correct, since only EIDE supports more than two drives
on a single device). According to Tony Sanders <sanders@bsdi.com>, a
patch has been created which allows use of extended IDE drives.
If interested, contact support@bsdi.com for more information.
I presently have two 1.8GB Quantums and will probably buy a 9.0GB
Seagate in the next month or two. The main reason is my ambition to
hold 30 days or more of news and thus be a definitive place to look
up previously created answers to questions and what-not. Because INN
wasn't designed to work well with a news spool split between
machines, I'm pretty much stuck using large SCSI drives on one
system, and I suspect most others will be too. For the record,
after reading Karl Denninger's comments on PCI's unstability, I have
postponed my Pentium/90 PCI + 9GB drive purchase, which I wanted to
do together. I may consider a Pentium system with EISA in coming
months. Recommendations along these lines would be warmly welcomed.
According to Tony Sanders <sanders@bsdi.com>, "Things seem to be
better nowadays wrt PCI. Buslogic finally seems to have stable
firmware for the 946. It's probably worth trying so long as you go
with a vendor that you know will let exchange it if things don't work."
His hardware recommendations are included later in this FAQ, under
"Tony Sanders' Recommended Equipment List".
Netcom gives each user 5MB of disk space a month, and charges for average
usage above that figure. So a gigabyte partition for users will only
last for about 200 users, and that only if you strictly enforce quotas.
Eric Raymond <Esr@locke.ccil.org> notes that you can get away with a lot
less if you use a customized BBS-style interface. "At CCIL, most of
our accounts go through a custom BBS interface; the users don't have
shell access. They get along happily on 300k each. Whether you can get
away with this or not depends strongly on what proportion of your users are
programmers."
Mark E Mallett <mem@mv.mv.com> notes that 300k "won't even hold a
.newsrc file. (Mine right now is 645k)." I suspect the difference is
that Eric's users may not read news in any great volume, thus keeping
the number of .newsrc files down. Mark continues: "We impose quotas
to help remind people to keep things clean. We initially started out
with no quotas; our disk space ran out within the first couple of
weeks. Added quotas and got at least 90% of it back."
My own system runs a custom-written newsreader that doesn't have a .newsrc
file; it works with a list of newsgroups the user has entered. I've
found that most people have extremely small groups files; a typical
user directory (which I just checked) takes around 6k. The main key
to this performance is that I let people "browse" through all groups I
carry without putting them in the groups file; only groups that have
been read at least once go there. In Tin and other readers, every
group's status goes into the .newsrc file even if I've never read it,
since I want to be able to see all groups on the list (even if I never
enter more than a very small fraction of them).
I would figure on about 300MB for the operating system, 1GB for users and
2GB for news, meaning that your disk requirements should be around the
3.5-4GB mark. Eric Raymond (quoted previously) seconded this motion,
only noting that it should support significantly more users than I wrote.
I suspect the key to this is really FTP. I don't allow FTP yet on my
system, and won't until I have written a customized version that
automatically downloads files to the user's system. That should solve
most of the common problems with disk space.
[NB Has anyone done this yet? I'm having a hard time getting the
time for this particular venture, and my users are just salivating
(you can see it, honest! :-) ) for FTP ...]
+.# Tony Sanders' Recommended Equipment List
Tony Sanders, founder of the Inet-Access mailing list, has compiled
some hardware information for use with BSDI systems. Much of this
should also apply to Linux as well, although (to my knowledge), Linux
does not yet have support for routing cards.
-- Here's Tony's document, reproduced with minor editing: --
These specs are intended for users building a high-performance,
Internet-ready PC fileserver or workstation class machine using BSD/OS
from Berkeley Software Design, Inc. Of course, users building more typical
configurations should also find it helpful as it lists resources for some
of the harder to find items.
This does not contain a complete list of supported hardware; please contact
info@bsdi.com for details on the operating system, a complete supported
hardware list, or with any questions you may have.
OS:
BSD/OS V1.1
BSDI World Headquarters
Berkeley Software Design, Inc.
7759 Delmonico Dr.
Colo. Spgs., CO 80919 USA
Toll Free: +1 800 800 4BSD
Phone: +1 719 593 9445
Fax: +1 719 598 4238
Email: info@bsdi.com
CPU:
486DX2/66 or Pentium
BUS:
ISA/EISA/PCI/VLB
For applications that are mostly network or CPU-bound a system with
an ISA bus will work fine, especially if you toss in a VLB disk
controller. You can get a perfectly good system this way and save
yourself some bucks.
Those with a bit of daring, after the cutting-edge in performance,
will probably want to try PCI. There have been some initial problems
with PCI but it is possible to get a working system with a bit of care.
Billy at ASA Computers can help you. Rumour is that most of the
initial problems with PCI 2.0 have been resolved. As always,
buyer beware (and shop with a vendor you can trust).
A good resource for all kinds of hardware (and who knows what else):
ASA Computers
+1 408 496 6853 [ext 201 -- Billy]
+1 408 988 0359 (fax)
+1 800 REAL PCS
Multi-Port Board:
Digiboard PC/Xem [~$1400 for 16 port card], can go to 64 ports.
RISCom/8 is a good price/performance alternative for less demanding loads.
[Note from the FAQ maintainer: For Linux, check out the Boca 2016
16-port serial board, which works fine on my system].
Router Card:
With the RISCom/N2, RISCom/H2, and RISCom/N1 synchronous cards you
can connect your BSD/OS directly to your 56K or T1 line and avoid the
extra cost of an router; probably saving your company thousands of
dollars! BSD/OS supports both CISCO HDLC and synchronous PPP framing
for compatibility with most service providers. Frame Relay support
is being worked on right now.
RISCom/N2 (56K to T1) -- Connects via V.35 to a standard CSU/DSU
RISCom/N2-S (approx $500) -- single port
RISCom/N2-D (approx $695) -- dual ports (very nice for routing hubs)
Mention BSDI when ordering to recieve special rates for BSDI customers.
Prices are approximate and subject to change.
The RISCom/N1 and RISCom/H2 cards are supported at 56K.
Available from:
SDL Communications Inc.
130 Liberty Street
Bronkston, MA 02401
+1 508 238 4490
For information on Internet Access Providers see:
http://akebono.stanford.edu/yahoo/Business/Corporations/Internet_Access_Providers/
ISDN is not yet directly supported (and we have not identified a
target card yet) but several people have reported using the
Combinet ISDN modem (120Kb/second):
Combinet
333 West El Camino Real, Suite 240
Sunnyvale, California 94087
+1 408 522 9020 (voice)
+1 408 732 5479 (fax)
ISDN BBS: telnet combinetu.combinet.com (login: isdn)
For more information about ISDN see:
http://www.crimson.com/isdn/
http://www.crimson.com/isdn/vendorinfo.html
http://www.icus.com/
http://alumni.caltech.edu/~dank/isdn/
Ethernet:
3COM 3C579 Etherlink III -- EISA
3COM 3C509 Etherlink III -- ISA
TNIC 1500 Transition Eng Fast ISA busmaster DMA NIC
South Coast Computing Services, Inc.
PO BOX 270355
Houston, TX 77277-0355
Email: info@sccsi.com
+1 713 661 3301
+1 713 661 0633 (fax)
Firewalls:
Setting up a firewall is recommended by many network experts
for additional security for your site.
The ``screend'' packet screening software package is available from
ftp.vix.com:pub/vixie/screend*. This package can be used to build an
IP firewall using your BSD/OS system.
You can get ``fwtk'' (firewall toolkit) from ftp.tis.com:pub/firewalls.
TIS also makes a commercial version called Gauntlet:
Trusted Information Systems
3060 Washington Road
Glenwood, MD 21738
Email: info@tis.com
+1 301 854 6889
SCSI:
Adaptec 1740/1742 EISA SCSI host adapter (Minimicro +1 800 275 4642)
BusLogic (BusTek) BT-946C PCI SCSI host adapter (w/firmware 4.21 or higher)
BusLogic (BusTek) BT-747A EISA SCSI host adapter
Disk space as required (Fast SCSI-II). 500MB-1GB is a reasonable
starting place for many configurations.
Corporate Systems Center
1294 Hammerwood Ave.
Sunnyvale, CA 94089
+1 408 734 DISK
+1 408 745 1816 (fax)
[Good resource for drives and other peripherals, new and refurb]
A full Usenet News feed will eat a *lot* of disk space (at the very
least 2GB) as well as disk performance. You should use multiple
smaller disks instead of one big disk to increase disk throughput and
minimize seek times (probably 3-5 2GB disks depending on how long
you want to keep stuff around).
Double or Triple speed SCSI CDROM (you can get a cheap drive if you
are just going to use it for installation). If you intend to really
use it you'll want a drive that supports SCSI disconnect or else it
will impact SCSI performance.
8mm Exabyte or 4mm DAT Tape Backup
Consider a Magneto-Optical system for site archives.
Video For Workstations:
Xstones Chipset Video Adapter
* 450,000 ??????? Number Nine -- #9 GXE128
* 350,000 MGA-II Matrox MGA Ultima
* 220,000 MACH64 ATI Ultra Pro Turbo (MACH64)
???,??? MACH8 ATI Ultra Pro (MACH32)
150,000 MACH8 ATI Ultra (MACH8)
* Supports resolutions upto 1600x1200
Others support resolutions upto 1280x1024
BBS/Menuing Software:
ftp.solinet.net:/pub/src/menu.tar.gz (src dist.)
ftp.solinet.net:/pub/src/unixmenu.SCO.tar.gz (BSD/OS src/bin dist)
ftp.solinet.net:/pub/src/yum.tar.gz (BSD/OS src/bin dist)
Other possible sources:
Eric Raymond's? Menushell? UniBoard BBS by <pizzi@nervous.com>??
TNSDrive by Vladimir Vorobyev <bob@fagot.turbo.nsk.su>:
ftp.turbo.nsk.su:/pub/unix/drive045-BSDI.tgz
[if you have any suggestions for this section please send them
to sanders@bsdi.com]
News:
INN (available from ftp.bsdi.com:contrib/news)
PageSat (Usenet News by Satellite) +1 415 424 0384
approx: $565 equipment, $30 a month (1 year contract).
One person on the inet-access mailing list (see below) commented:
:: The chief reason there are holes is that PageSat is delivered via
:: satellite. When it rains hard, we lose the signal, and they don't
:: retransmit any data. The other big problem is that they only have about
:: 100MB/day of bandwidth (synchronous 9600bps == 1200cps). Usenet is
:: currently around 130MB/day, so they have an obvious lag problem.
:: Even with these problems, PageSat is worth it because it eases about
:: 100 MB/day of bandwidth off of my lowly 56k circuit.
Note that a 56Kbps link is ~590MB/day of bandwidth. Of course you
aren't going to get that in practice and during peak usage things can
get pretty bad, so offloading 100/MB of real work could be a big win.
SLIP/PPP:
Basic SLIP/PPP client/server code included with BSD/OS. For a more
advanced setups you might want to check out Morningstar PPP:
Morning Star Technologies Inc.
1760 Zollinger Road
Columbus OH USA 43221-2856
Email: Marketing@MorningStar.Com (sales e-mail)
Email: Support@MorningStar.Com (technical e-mail)
FTP: ftp.MorningStar.Com:pub/
WWW: http://www.MorningStar.Com/
+1 614 451 1883
+1 800 558 7827 (Toll Free USA and Canada)
+1 614 459 5054 (fax)
Internet Access Providers Mailing List:
This is a mailing list for people to discuss issues about being
an Internet Service Provider. The list is not BSD/OS specific.
EMail inet-access-request@earth.com to join the list.
Things to consider:
Administration machine
DNS, News, telnet, FTP, gopher, WWW
Fileserver
Firewall
Shell Accounts
Modem Pool/Terminal Server
Configurations for Mac/DOS/Windows clients:
Draper Kauffman, Ed.D. <draperk@io.com> wrote on inet-access:
:: The Internet Starter Kit for Macintosh, by Adam Engst
:: Hayden Books, $29.95 (ISBN: 1-56830-064-6)
:: The Internet Starter Kit for Windows, by Adam Engst, et al.
:: Hayden Books, $29.95 (ISBN: 1-56830-094-8)
::
:: Besides being excellent introductions to the Net, they include all the
:: software a user needs to set up a SLIP or PPP account with email, news,
:: ftp, telnet, gopher, web, finger, etc., etc., running as client software
:: on a Mac or a WinPC.
-- End of Sanders document --
+.# What equipment is needed to hook up my system to the Internet?
To try starting an ISP, all you really need is a 14.4kbps or 28.8kbps
modem and a resellable SLIP connection to another provider. This is
my current situation, as I try to gauge receptivity to my ideas and
tune up the software. Unfortunately, resellable SLIP may be difficult
and/or expensive to obtain.
Aaron Nabil of internetworks (i.net) was kind enough to write a
response with some interesting comments about resale. Actual resale
policies vary depending on the vendor you select. "If you call one up
and ask if they permit 'resale', to which they answer no, you
shouldn't go away thinking they can't help you." Perhaps they can.
Examples of different policies, from Aaron's message:
* ANY TIME you make money from the connection, even if you just
charge people to access a Gopher or WWW server, or sell products
through that server. (Seattle providers)
* Selling shell access to a computer connected to them is resale
(Netcom). (He actually said "Netcom?", but I know from asking
them that this is indeed the case).
* Selling permanently addressed IP is "resale", transient IP isn't
(Possibly Alternet)
* Selling any IP is resale, but shell access isn't. (Internetworks
[i.net, his company], most other providers
* We don't care what you do. (Sprint, free-nets, ISA Network Associates).
(I added ISA because I happen to know this is their policy).
So, if you find a vendor and discover that they are not amenable to
resale of their connection, make sure you know the precise definition
thereof. You might be pleasantly surprised. You may also want to
ask if they can "work something out"; special arrangements are possible.
The next step above SLIP is a direct connection at a 56kbps data rate
(double the speed of a 28.8 modem). The TLG Leased Line FAQ
(ftp://ftp.tlg.org/pub/FAQs/tlg-leased-line.faq) explains why a 56k is
significantly better than a 28.8 SLIP; it has to do with latancy rates
and other neat concepts I don't remember. I will later summarize
portions of that FAQ here.
To deal with more than a very small number of customers on a paying basis,
you will need a 56k or T1 connection hooked up to an internal network.
For this, you need the following equipment:
- A Router. This is a box that hooks up to your local network and sends
out packets destined to the Internet, while leaving your local packets
in your local network. You can theoretically program a PC (using BSDI
software) as a router, but the rather vague impression I get is that
this is for true experts only.
Aaron Nabil of i.net comments: "Well, it's not necessarily for experts,
but don't expect your carrier to help you fix it if it breaks."
Tony Sanders <sanders@bsdi.com> adds: It's not really that hard and
one advantage of using a system for the router is that something
like a 486/66 or Pentium can pull double duty for a small provider
with an eye on expenses and serve as a mail/www/gopher/ftp/news
server at the same time as being the router (easily up to T1 speeds).
Of course, as you move up in the world you might need to move to a
dedicated router but I think that in general it's overkill for
someone just getting started. For more information on BSDI
hardware, see Tony's guest contribution, above.
On the question of why Linux or other free Unix systems can't be
used in the same way, Tony writes: "I think the problem is that they
don't have support for any interface cards that talk to a CSU/DSU.
Of course, that may have changed." Some inventive Linuxer is
bound to change this eventually, but until then that seems to
give BSDI a strong edge.
Further information on the subject of using a BSDI box as a router
is included at the end of this FAQ as Appendix A due to its extraordinary
length (circa 400 lines).
i.net is one of the few providers with the guts to list actual costs
associated with the connection, including equipment. They charge $ 2,200
for a Cisco 2501 router. It can route packets up to the T1 level,
and should do fine for most providers. "The Cisco is the benchmark
of routers, the reference by which other routers are judged." Its
main limitation is that it has only a single ethernet port and
two serial ports, so it's pretty much limited to going from your
internal network to the Internet. More expensive routers, like the
Cisco 7000 series, can handle several such concurrent connections,
and thus pass packets between several different internal networks.
For almost all start-up providers, the 2501 should be sufficient.
[I believe there are now routers that also serve as terminal servers;
hopefully someone who actually knows something about hardware can
contribute something on this vital subject].
For $ 1,700, Internetworks will also sell you a IRX-11, which will
also handle 56k and T1 connections. "It's just not as nice a router
as the Cisco."
Net-99 is giving a 30% discount on CISCO routers and other equipment
if you become a customer. A 20% discount is available from them for
non-customers.
- A CSU/DSU. About $ 550 (again, using i.net's price sheet) for 56k or
$ 1,300 for T1. This is the equivalent of a modem - it translates
the router's output into a signal that can zip through the telco's
lines.
Aaron: There are cheaper brands for 56k, such as Adtran or Bat, that
sell for about $ 250. "Try to get a 56/64k CSU/DSU if possible."
Sean Shapira <sds@jazzie.com> writes, "I have no complaints regarding
my BAT 56k CSU/DSU."
- A local area network to connect your router to your other computer(s).
This could be an ethernet card in your system and Ethernet connections
or 10Base-T, involving a hub and other stuff. (Someone else might
want to flesh this out; this was modified through suggestions from
Aaron Nabil and Sean Shapira). If you use your sole BSDI box as
a router, you might be able to skip this, but most providers with
a T1 or greater need a network to split their load between multiple
machines.
- Cables to hook everything together. They're easy to forget, but,
according to Aaron, "they are non-trivial!"
Depending on your specific Internet provider, you may need to provide
the router and CSU/DSU at (a) your own site only; (b) at your site and
their site. Some connections, called "full service", will provide
the equipment and maintenance at both sites; a good example of this
is CERFNet. This can be contrasted with ISI Network Associates, which
requires you to provide the equipment on both sides. Naturally, ISI
is a LOT cheaper than CERF. However, for a 56k connection, i.net is
cheaper even than ISI, and they provide the router on their end.
Go figure.
+.# What other equipment do I need?
To start receiving calls from users, you need modems, telephone lines
and one or more multi-port serial cards or terminal servers.
[NOTE: The rackmount modem section is in development and this may contain
some inaccurate information].
MODEMS: Many people swear by rack-mount USR or other big name modems.
You can do all sorts of neat things with them, such as remote test and reset
of individual lines. Unfortunately, they cost about triple what
low-end standalone units do. Low-end standalone units are more likely
to not connect properly after a disconnection. This appears to be
especially true of US Robotics Sportster units, although their high-end
modems are superb. I've had excellent success with Intel 144e modems,
which cost all of $ 99 each. One note on the Intels: They have
apparently changed their design significantly in the last few months.
The newer ones are instantly recognizable by their roughly square
power bricks; the older ones were clearly rectangular. The difference
relevent to Internet providers is that the older ones support "at&q6"
to turn on error correction and autobauding. The comparable
command for the new series is "atb0\j1". The \j1 turns on autobauding
and the b0 specifies CCITT mode. Note that the "\j1" will have to
be typed in as "\\j1" on most Unix systems. Unfortunately, Intel
is apparently getting out of the modem business. I, for one,
will miss them.
Alicia Salomon (salomon@seas.gwu.edu) was kind enough to send me a
price list for US Robotics rackmount modems. I have to say they
seemed pretty forbidding, even with special "Internet Pricing". The
basic chassis, including "dual power units and the Network Management
Card Set (NIC/NAC)" is $ 3,810. From what I gather, this gives you
the ability to reset and reassign modems remotely by connecting to the
modem's ethernet slot and giving them commands. You then must buy a
Dual T1 card set for $ 2,701 to hook these modems to a T1 line, which
can then be brought in to your network. A modem card, which contains
four modems, costs around $ 2,000 (again with the special pricing).
So if you wanted to start with 16 lines, it would cost a eye-popping
$ 14,511; 16 Intel 144e modems would run only $ 2,240. Filling the box
to its 48-line capacity would cost $ 38,511, while 48 144es would
cost $ 6,720. However, this is not quite a fair comparison, since
this price apparently includes the equivalant of a terminal server,
which would otherwise cost somewhere around $ 2,000 for 16 lines.
Despite this, I suspect the rackmounts have their place. If you have
a POP that's remote from your main business location, you might not be
able to go there and physically reset the modems in any reasonable
period of time. With that situation, rackmounts might actually be the
best solution, since you'd just reset them via your network. An
alternative would be to build your own remote switching device, so you
could remotely switch the modems on and off when they needed to be
reset. Even hiring someone to design and build such a thing might be
cheaper than a rackmount modem box.
Sean Shapira <sds@jazzie.com> notes the primary advantages of
rackmounts: "space savings; significantly reduce wiring harness".
These are undoubtably true, and I suppose I'd pick up a nice rackmount
modem if I had an unlimited budget. Pity nobody does in these tough
times. :-(
Eric S Raymond <esr@snark.thyrsus.com> took the opposite approach in
his system setup, which I think will appeal a great deal more to the
startup ISP. His contribution also gives some us some idea of the
kind of patching BSDI users have had to do. He writes the following
paragraph about his experiences:
"We got our nonprofit ISP started using a super-cheap modem
called a LineLink 144e, built around the Rockwell data-pump chip (same
one used in the Zoom and Boca modems) and costing $105. We hooked our
modems to SDL RISCOM/8 multiport boards, the brand recommended by
BSDI. This setup has worked pretty well, except for one major problem
--- when UNIX on our 50MHz box hangs up, the DTR-low interval goes by
so fast that the LineLink sometimes fails to see it. This causes the
modem to hang in the off-hook state, blocking the line and requiring a
manual reset. I worked around this by patching a 250msec delay into
the RISCOM driver's DTR-pulldown code. This fix may become
unnecessary when RISCOM releases the next driver version, which is
supposed to do true hardware handshaking on the modem lines."
Incidentally, the setup I have under Linux, featuring a BocaBoard
2016 16-port serial card and $ 139 [now $ 99] Intel 144e modems, has
worked flawlessly with no installation problems at all.
MULTI-PORT EQUIPMENT: Your modems have to connect to your computer,
which normally has two or fewer available serial cards. How to do
this? There are three basic ways:
DUMB CARDS: These cards give all processing to your system's CPU,
which makes it run slower than the alternatives. However, they
are very cheap and relatively easy to set up. If you have a
Linux system, you should request my BOCA-FAQ, which outlines
the procedure for setting up a Boca 16-port board under Linux.
SMART CARDS: These cards take some of the processing load off
the main system. They are supported by the BSD systems, but
not Linux; as a result I have limited knowledge of them.
TERMINAL SERVERS: These are high-end products that are mercilessly
expensive, often over $ 2,000. In return, you get a device that
handles your terminal ports by effectively telnetting to your
system. This is significantly easier on your system, since it
no longer needs to process any form of terminal interrupts. This
is, however, yet another item I could use some help on, since
I've never used one of these beasts - too much money!
Karl Denninger <karl@mcs.com> writes the following on various brands
of terminal servers:
Annex:
Unix-style kernel. Well-known, venerable, highly stable.
Decent performance, but you will NOT be able to drive all the Annex
ports to full speed at once. I've tried it.
Livingston:
Newer, good reputation, but has a few problems that I can't live
with (primarily no host route advertisement). They claim to be
fixing this. RADIUS authentication system is quite nice. Not a
bad box.
Telebit:
Venerable, well-understood, *extremely* flexible command set and
capabilities. Can handle leased connections as well (up to T1 with
appropriate cards) which makes it a "POP in a box" possibility.
Classics are available cheaply, current units (NB40s, etc) are more
expensive. Will route IPX and Appletalk in addition to IP.
Basically it's a question of what you want and need.
+.# How many phone lines do I need?
To start, it pretty much depends on your budget. I currently have
four: three incoming lines plus my SLIP connection. Because my system
is experimental, nobody calls it yet and so I have no hard answer to
this question. (After a few months of running software that works
reasonably well (not fully debugged by any means), my lines are now
often full). My estimate is that you want 8-10 lines to start,
once you're ready to give your system a bit of publicity. But it
really all depends on your market and how high a profile you can
maintain.
Since I've written this, I've started getting occasional busy
signals on my three-line system. I have about 100 user accounts
and 10 people who call several times a day. Since I don't charge
for the system yet, however, most of these numbers are meaningless.
Stay tuned.
As a general rule, 10 users per line is suggested for conventional
dial-up connections. I believe Karl Denninger maintains roughly
this same ratio, even with his SLIP connections; he can do this due
to a 20-minute idle timeout for the SLIP.
Alicia Salomon <zone@access.digex.net> adds, "after about 400 users,
it goes to about 12:1 and then goes to 15:1 around 1000ish (guesstimates
based on vague sources of data input)".
Permanent SLIP connections by definition take precisely one dial-up
line per user, and should be priced accordingly. Some people have
gone to 4-6 users per line even for non-permanent SLIP.
*.# What about Residential phone lines versus business?
Residential phone lines are generally much cheaper than business
lines. Traditionally, the phone company has subsidized the cost of
residence lines by charging businesses substantially more for phone
service. Fortunately, the network provider has almost exclusively
incoming calls, so the measured service issue for business lines
doesn't come up much. However, the base rates for residence lines
are still about half the cost of business ones.
Whether you use residence lines or not obviously depends at least in
part on the location of your business; you cannot use residential
lines if you're in a business location such as an office building or
store. Phone company policies on use of residence lines for a
business operating out of people's homes vary dramatically depending
on the area in which you live. The worst case is that you could be
back-billed for business rates from the inception of your business.
Best case, of course, is that the phone company could ignore you
completely and allow you the residence lines without trouble.
Fortunately for the budding ISP, phone companies in many cities are
setting up telecommuting programmes and in general trying to
encourage people to work at home. One of the major elements of this
appears to be allowing home businesses to "get away" with the use of
residence lines. Because of this, the pressure to get business lines
seems to have abated in many areas. For example, a few years back,
Pacific Bell tried to institute a crackdown against people who were
using residential lines for their BBSs. However they have recently
reversed this policy; now, they are more than happy to take orders
for multi-line BBSs at residential rates; my rep knew exactly what I
was doing when I placed the order, and even offered a few words of
encouragement. When I asked if there might be trouble in the future,
my rep - who seemed like quite a knowledgeable fellow - said no.
According to Alan Byrant's book on running a successful BBS,
Southwestern Bell has been particulary aggressive at nabbing
providers and BBSs who try running business systems off residential
lines. (I read this in the bookstore and unfortunately don't remember
the name of the book).
+.# Where do I put all this stuff?
Most beginning ISPs start in their owner's home. This is nothing to
be ashamed of; even mighty NETCOM started this way. Of course I'm
just a shade biased here; my nascent provider is right here at home,
too.
The home address has some very interesting advantages:
(1) It doesn't cost any more than what you're already paying for rent
or mortgage.
(2) You can use residential phone lines instead of business; rates are
50% or less business costs. However, see above for a full
discussion of the issues behind this.
(3) It's easy to get to in case of an emergency. Just walk to your
desk.
It also has some disadvantages you might not be aware of:
(1) It's probably technically illegal, thanks to zoning rules. Fortunately,
these are being gradually loosened.
(2) The tax consequences of deducting the portion of your mortgage used
for business purposes are extraordinarily murky - and whether you
rent or buy your home, office in home deductions are a major red
flag for IRS audits.
(3) It has a few image problems. You don't normally want to escort
potential customers for big accounts into your living room and
have them run into that strange mess the kids created during
playtime. Potential employees may also be less than impressed
by your working environment.
(4) Some telco services may not be available to residential customers.
This is one issue I don't know much about - I'd appreciate more
information from people who know what they're talking about here.
In the end, though, it all boils down to money. If you have tons of
the stuff, you'll probably have an office. If money's tight, running
your ISP out of your home is one of the best ways to save.
+.# This is so expensive! Where can I find this stuff cheap? (*)
Don't forget the newsgroups misc.forsale.computers.workstation and
the corresponding pc-clone groups for the best prices on hardware.
Unfortunately, much of what's offered is rather low end and not
really suitable for an ISP. I did get my Sun 3/60 through the
workstation group, but it's not going to power a full provider; I
use my Linux PC for that and use the Sun for its neat 19" colour
monitor.
If you're looking for PC-style hardware to run your provider,
Computer Shopper is one of the best sources of deals.
For Sun workstations, good places to search are ... [here's a neat
place for other people to fill in].
Karl Denninger <karl@mcs.com> is selling routers and CSU/DSUs at a
substantial discount. Naturally, he'll give you a better deal if you
decide to become a Net-99 customer. See below for information on
Net-99.
Henry Minsky (hqm@ai.mit.edu) is developing a list of dealers who
carry used equipment. Access it through a Web browser via the URL
http://www.ai.mit.edu/datawave/hardware.html . "There are some
pieces of equipment I would not recommend getting used, but
others, such as the tape drive, have warranties from the reseller,
and seem to be a good option."
+ Hooking up to the Internet
+.# How do I hook up to the Internet?
By finding an Internet provider in your area who's willing to hook
you up, or by connecting with the major services such as Net-99, SprintNet
or AlterNet.
+.# What are the levels of connection you can buy?
Maximum
Connection Equipment Required Approx Cost Simutaneous Users
28.8K SLIP 28.8k modem $ 300 around 3
56k Leased CSU/DSU 250 around 8
Router -or- routing card 2300 or 500
T1 Leased CSU/DSU 1400 around 200
Router -or- routing card 2300 or 500
Some connections will require that you buy a modem, router and
CSU/DSU for both ends of the connection. ISI Network Associates, for
example, requires this, meaning that their 56k connection would cost
their $ 1,000 startup fee plus two times $ 2,550 - a total of $
6,100. As an alternative, you could pay a $ 4,000 startup fee and
still buy the equipment on your end. Other connections furnish the
equipment on both ends, including a service contract. Unfortunately,
the latter type of connection is normally prohibitively expensive.
The telephone company rates for these connections range from reasonable
to horrendous. On the low end, the phone line I use for my 28.8 SLIP
is a conventional residential line which costs around $ 20/month.
Leased line fees for a 56k line range (in Southern California, anyway)
from about $ 100-200/month. T1 fees range from $ 400 - $ 1,200 a
month, depending on the distance between your site and the nearest
provider POP.
+.# What is Frame Relay?
The following Information about Frame Relay connections was
contributed by Sean Shaprio <sds@jazzie.com>; unfortunately my
explanation is sufficiently changed from his original words that
errors are particularly likely. He suggests that you read "the
little 30 page book (published by Motorola?) that describes it
in excellent detail" if you are seriously considering this.
Frame Relay connections are an up and coming form that are closer in
nature to a switched telephone connection than a traditional 56k/T1
link. To connect to a provider with frame relay, you run a leased
line to the nearest frame relay access point. The connection is then
made to your provider, even if it is a very long distance away. The
provider runs a high speed connection to his nearest frame relay node,
where it can then get split off to several connections. So instead of
having multiple 56k connections to his customers, he pulls a single
T1 into a frame relay switch; the packets are then switched over to
the customer's 56k hookups. This is the service that lets providers
like Netcom and CRL operate nationally while still having all their
equipment in their original Bay Area headquarters.
The main problem with frame relay is that you are sharing a switched
line with a potentially large number of other users; this line has a
fixed capacity that is divided between you and them. Your provider
will give you a committed information rate (CIR), which tells you the
minimum speed your line will connect at, no matter what the
conditions. Sprint presently gives a CIR of zero, which means that
they do not guarantee that your connection will be continuously
running at any particular speed, or even operational at all. So in
theory, if you were running a Frame Relay T1, you might have a
virtually worthless connection if all the other connection users took
up all the bandwidth.
The advantage of Frame Relay should be fairly obvious; since you're
sharing a large connection with other people, you aren't paying the
phone company for hideously expensive leased lines. Because of this,
Frame Relay is a much cheaper service than the traditional 56k or T1
leased line.
This form of connection requires a special frame relay compatible
router to work.
This explanation could be wrong; I'm still not sure if I fully understand
frame relay. Additions and criticism to this section are welcome.
Some questions for the experts on this list:
* What are frame relay provider (telco) rates like?
* How is the performance compared to standard 56k/T1? Karl
Denninger <karl@mcs.com> and his partner Joseph Stroup initially decided
against offering Frame Relay with their Net-99 service due to low
connection quality. However, they are currently offering a pilot
Frame Relay project at an attractive rate; they will expand it if
it meets their quality standards.
+.# What are the main national providers, and how much?
For T1 access, your cheapest national provider is probably Net-99.
Karl Denninger of Net-99 <karl@mcs.com> picked up the following
representative rates for T1 access: $ 1,999.00 per month for Net-99;
$ 2,700/month for Sprint and almost $ 7,000/month for Alternet. The
latter price is their special (high) rate for resellers; Sprintnet
and Net-99 both allow unlimited resale without restrictions. A
number of people have commented that AlterNet is apparently no longer
selling to resellers; AlterNet representatives have denied this.
Their rate, however, appears to take them out of competition in any
case.
Net-99 is too new to judge quality, although initial reports seem to
be favourable. I received one positive comment for Net-99 and no
negatives when I asked about provider quality on the inet-access
list. Most other providers seem to offer high-quality connections,
but exceptionally poor customer service responsiveness.
If the national provider rates seem too expensive, you may wish to
hook up with a local provider. Watch out for the connection quality,
though; if the local company sells you a T1, and all they have is a
T1 connection themselves, you're bound to get mightily poor throughput
to the rest of the Internet.
+.# Where could I get a list of national and local providers?
An excellent starting point is the DLIST, "an online list of Internet
service providers who offer dedicated line connections." To find out
how to receive an updated version of this list, send mail to dlist@ora.com.
(From the book 'Connecting to the Internet: An O'Reilly Buyer's Guide',
by Susan Estrada, which includes a printed copy of the DLIST in an
appendix).
Here is the ``Yahoo'' reference, which is apparently more up to date:
Business:Corporations:Internet Access Providers
http://akebono.stanford.edu/yahoo/Business/Corporations/
Internet_Access_Providers/
Note that this all has to be on one line.
+.# What about a SLIP connection?
For the most part, a SLIP connection is not considered sufficient
for a serious provider. However, it may be the best way to start if
you are unsure of demand for your service or want to test-market
your ideas. It certainly lets you hook up for a minimal amount of
money, assuming you can find a resellable connection.
If you get SLIP, try to get CSLIP (compressed SLIP), and make sure
you know what baud rate the line is fixed at. I got a 28.8k
SLIP with the baud rate fixed at 28.8, and the result was that
I could not get a newsfeed consisting of alt.* plus rec.* without
falling behind. Be sure you get as good a SLIP connecion as
possible.
+ What are interchange agreements, and why should I care?
+.# What is the Commercial Internet Exchange (CIX)?
The Commercial Internet Exchange (CIX) is many things to many people.
To some, it is the heroic institution separating small ISPs from the
oblivion of disconnectedness. To others, its $ 7,500 a year annual
fee is restraint of trade, monopolization or worse.
The CIX offers to do three basic things for you:
(1) Any CIX member must agree to pass through the packets of any other
CIX member, without fee. So if you want to talk to a site in England,
and you're going through five or ten other sites on the way, they
cannot charge you for this transport service, assuming all the sites
are CIX members. (Karl Denninger <karl@mcs.com> tells me in response
to this document that they are only obligated to connect through the
CIX router and nowhere else, although you are presumably guaranteed
at least your access to the CIX router).
(2) The CIX operates a router, based in San Jose, CA. As part of your
membership, you have to be connected to that router either directly or
through a chain of CIX members. If you cannot connect to a specific
site through normal means, you can use the CIX router as a last-resort
option to get where you need to go. A direct connection to the router
costs $ 5,000 on top of the membership charge.
(3) Unspecified lobbying and public relations efforts. Observers on
the inet-access mailing list state that these efforts are negligible.
However, those people are also anti-CIX for other reasons, so you may
wish to take their opinions with a grain of salt. The CIX has not
issued any official commentary that I know of on the specifics of
these efforts.
There has been an enormous amount of talk - and we're talking about
literally megabytes of stuff - about whether the CIX as it stands is a
Good or Bad thing. Non-CIX members and some CIX members not on the
board have made the following points:
(1) The basic principles of the CIX are worth saving. Even the most
virulent hater of the organization has stated that the principles of
settlement-free peering are the glue that holds the Internet together
- and those principles should be supported by a trade organization
people join. Gordon Cook's <gcook@tigger.jvnc.net> remarks are
typical: "CIX may well go away, but if it does Sprint and MCI in a
year will have life and death power over the small fry ... etc ... ie
what if they forbade resale?" Gordon is no friend of the CIX in its
present state; he has made many scathing commentaries on the present
situation.
(2) The CIX is imposing route filtering, as of 15 November 1994. Previously,
if you were a non-CIX member and were connected through a CIX site, you
could still use the CIX router in the same way CIX members could. Under
route filtering, the CIX router will become for the exclusive use of
CIX members. Non-CIX members consider the $ 7,500 annual fee to be
excessive; depending on the size of their operation, it could nearly
double their operating costs.
In September 1994, the membership had an annual meeting, at which
they voted to NOT impose this route filtering. The CIX Board of
Directors said that they would have to impose filtering in any case,
for legal reasons. Some members were relying on the CIX connection
to hook themselves to all providers, whether members or no, so they
actively did NOT want the filtering to occur. At the end of the
meeting, people like Karl Denninger <karl@mcs.com> were confident
that filtering would not be imposed. However, on November 1, the CIX
announced that, for legal reasons, the filtering would be imposed on
15 November. At that point, Karl Denninger's provider and Net-99, a
joint venture of Karl Denninger and Joseph Stroup, made a very public
resignation from the CIX. See below for some additional details).
On 15 November, supposedly D-Day for filtering, someone on the
Com-Priv mailing list noted that nothing had happened, and that even
CIX routes were still available to all. Bob Collet
<rcollet@sprint.net> admitted that the filtering was being
implemented only gradually. Some people have said that filtering is
impossible, given the specifics of route handling in the type of
router being used by the CIX.
(3) The CIX has a router, and some people are a long, long way from it.
Wouldn't it be better to have lower membership rates than a router,
especially since the CIX can be thought of in some sense as competing
with its own members? (In fact, one of the CIX members - PSI - operates
the CIX router under contract).
All this sound and fury can be quite entertaining, unless your
business depends on it. However, you may be able to relax: only 38 of
32,000 routes are unique to the CIX router. What this means is that
if you're not a CIX member, route filtering will do very little for
you; you lose contact with just 38 sites. If, however, you are a CIX
member relying on the CIX router to connect you to the world, you may
find yourself cut off from numerous non-CIX sites. As a result, most
of the people connected through the CIX are apparently scrambling for
alternative connections even as we speak.
What seems to have happened is that connectivity providers such as
Sprint and Net-99 are in practice taking care of routing for their
customers. As a result, the CIX router has become very nearly
worthless, especially since it's been heavily overloaded. Bob
Collet's <rcollet@sprint.net> latest statement is that "a phased
deactivation of the router" will occur, probably to conclude in late
1995. At that time, the CIX will then become a (cheaper) trade
association which will continue working for the settlement free system
as described at the beginning of this section. Other CIX Board
members, however, are hurling insults at all who would dare to ask
them questions.
The following comments have been made by people on the Com-Priv
mailing list on this subject:
* If you're connected by a provider that has access to MAE-EAST,
a major interconnect point, you'll be able to connect to people
hooked in to all major providers, since they're all there. This,
for example, should cover both Sprint and Net-99 customers well.
* However, you will NOT be able to connect to people who are
hooked in solely to the CIX router, unless you become a CIX
member. Fortunately, very few sites appear to be in this
category, and most of them are scrambling for alternative
arrangements even as we speak.
* Providers don't advertise all their interconnected routes to
MAE-EAST; they only advertise those routes which help provide
connectivity to their customers. For example, if PSI has a
direct link to Hong Kong, and PSI is connected at MAE-EAST,
you'll be able to hook up to all PSI customers, but not necessarily
Hong Kong. However - just to confuse things - if the Hong Kong
Supernet, a specific site, was connected through PSI, you would of
course be able to hook up to it.
As a result of this CIX filtering, Karl Denninger's MCS-NET and
Net-99, his joint venture with Joseph Stroup, have resigned from the
CIX effective immediately. Their complaint is that the obvious wishes
of the membership (as voted on in the earlier meeting) were ignored.
Bob Collet has asked them to suspend their resignations temporarily,
while he attempts to put together a solution to please all parties. As
a result, they have not yet put their resignations in writing, but their
very deep displeasure with the CIX and its representatives could
hardly be more clear. The CIX response has been that the filtering is
a legal necessity, according to advice of counsel. Others say that
the law could be read either way, and that the primary goal of the CIX
is to expand connectivity; filtering is unquestionably not going to
advance this goal.
It's worth repeating the requirements for filtering, to give us
a better understanding of what's going on. Filtering would affect
you if:
- The site that feeds you is not a CIX member
- You are not a CIX member -AND- you sell SLIP/PPP access, or any
other form of IP connectivity (56k lines, T1, etc).
Otherwise, you can relax and ignore this issue completely. So if
you sell shell or BBS accounts, and the site you're connected to
is a CIX member, then you have full CIX connectivity and can
cheerfully ignore this issue. Otherwise, the bottom line seems
to be that you could ignore it anyway. Stay tuned, though:
this optimistic vision is probably good for the next 30 days and
no longer.
Bob Collet (rcollet@sprint.net), a spokesperson for the CIX Board,
was kind enough to review the above. As his conclusion to a short
list of suggestions, he wrote: "Suggest toning down the personal
opinion flavor of the document." His belief is that my anti-CIX
prejudices are showing rather blatantly, and that an impartial
document would be more neutral in tone. I have invited Mr Collet or
any representative he wishes to choose to contribute a statement of
CIX policy, which I will insert here verbatim if and when it arrives.
Bob Washburn was Executive Director of the CIX until a month or two
ago. Mr Washburn was taking care of most of the public relations for
the organization, and his departure seems to have left the PR tasks
up to members of the board. They are presently searching for a
replacement, which may be one reason the flow of information out of
the CIX is so sluggish. There are now approximately 100 CIX members.
It is thus worth noting that this document consists of an analysis
created by me using the best information I know - the various
messages I read on the inet-access and com-priv mailing lists.
While I hope it has been of value, clearly it cannot guarantee
anyone's future policy, and does not constitute legal advice.
I will definitely plead guilty to making an attempt to give this
rather dry material a bit of entertainment value.
+.# What is NET-99?
According to Karl Denninger <karl@mcs.com>, one of the founding
members:
"Net99 is a commercial effort by Joe Stroup and myself to provide a
resale-encouraged, peering-mandatory, backbone environment for the small
and medium sized reseller. Net99 is not a cooperative, or a non-profit.
Net99 is, however, an alternative project which should, if preliminary
indications are good, provide a reasonable and affordable alternative for
the reseller of Internet connectivity when squared off against the large
companies now domainating this portion of the network infrastructure."
Net-99 beat their November 1 deadline by a very considerable margin,
having POPs up in mid-October. They presently offer service in the
following cities:
New York, Houston, Los Angeles, Chicago, San Jose, Boston, Philadelphia
When asked for the number of sites presently on Net-99, Karl said,
"Can't give you an accurate count; its growing incredibly rapidly!"
Robert Gibson <wa3pxx@cais.cais.com> writes us as follows:
"We have a connection via T1 to NET99, and it works very well. We
also have other T1 connections into the Internet via Sprint, Navy.
I have found the connectivity and support GREAT. ... NET99 is
*great* in terms of service, and I would gladly pay a few $$$ for
service, and enough $$$ to keep the network growing."
+.# What is Metering, and why are people so emotional about it?
Metering is the idea that users and service providers should be charged
depending on how much use they make of the service. At first blush, it
seems only fair; if you're on the net for an hour a week, you should
pay less than those who are on for five hours a day. For customers of
metered services, however, it tends to be a very bad idea indeed.
There are really two types of metering: Usage-based charging of users,
and usage-based charges made by backbone providers. Very few people
have much against the former, since free competition seems to be eliminating
it (other than the toll-free number options of many vendors, which have
to charge by the hour to pay the phone company's cut).
The latter, however, is a different kettle of fish. Most of the time,
when people talk about metering, they are talking about charges per
packet for Internet use. As a provider, you would be charged for backbone
use by your connection provider (such as SprintNet). The problem, of
course, is that this means you'd have to pass those charges to your
users in the form of hourly fees; this would effectively eliminate the
flat-rate pricing model that's been so successful.
What's the argument for flat-rate pricing? There are really three:
People are much more eager to use a resource when it's not metered.
Most people won't use a service for anything but the most vital needs
when they hear the clock ticking in their heads. The net's ethic
is founded on volenteer work of all kinds, like the production of
FAQs such as this one, the moderation of newsgroups, and so on.
This work would be prohibitively expensive with metered use, and
the amount of information available would be much lower.
Second, flat-rate is almost always cheaper for the user than the
metered option. Consider two pricing models I've seen:
$ 6/month plus $ 2/hour (UUNET)
$ 17.50/month flat rate (Netcom)
If you spend just 6 hours on the net each month, the flat rate becomes
cheaper than the metered price. If you really enjoy the net and
spent 3 hours a day on the system, your bill would be nearly $ 200
a month!
Finally, measured accounts can be a real hassle due to the difficulty
of administrating the timed accounts. This administration costs a
remarkably high percentage of the gain in revenues obtained.
The Internet has become the lively, fun and often bizarre resource it
is today largely BECAUSE people didn't have to pay by the hour or by
the packet. If the big communications near-monopolies ganged together
and offered only hourly rates, the Internet would be changed beyond
recognition. This is why people who are normally the most ardent
free market advocates are pushing for some kind of regulation of backbone
services.
+ Internet Software tips, tricks and answers
After looking at all these incredibly high costs associated with getting
on the Internet, it is somehow a relief to venture into the world of
software, where virtually everything you would need to use is available
at no charge other than a few hours of pain setting it all up.
There is a massive amount of information on the net on Internet
software and how to set it up. The O'Reilly zoo, a collection of
books for Unix system adminstrators with pictures of animals on the
cover, is strongly recommended by most people who've read them. I
have a pretty healthy zoo, and fortunately the cover critters haven't
chewed through my cables just yet. (I keep a close watch on them,
though).
Because much of this stuff is well known to anyone dropping by the
bookstore or reading their manuals, I'm only going to mention some tips
I've gotten through (sometimes hard) personal experience.
+.# The Permissions headache
Perhaps the hardest thing to get right when setting up your provider
software is Unix permissions. The permissions system is an indispensable
way to straighten out who can do what on your system, but it becomes
absurdly tough to manage certain programs. I managed to lose three
weeks worth of news while vainly attempting to straighten out some
particulary ferocious problems of this type, although this was mainly
because I was unlucky enough to have left down before checking that
one simple change I'd made would hold up. Don't repeat this mistake -
always check permissions whenever you make any change, however minor
they may seem at the time.
+.# Which news software should I run?
This is an interesting question, since most of the books and other
documentation on news software are way out of date - particulary
the O'Reilly/Todino book that is so often described as the definitive
guide. At the time that work was written, C-News was the classic,
recommended release of news software.
Because of that book's recommendation, I spent more time fighting
with the incredibly slow C-News unbatching software than I care
to remember. It would virtually freeze up my system whenever news
was being unloaded!
As soon as I switched to a SLIP, I switched to InterNetNews (INN).
What a difference! My system does not slow down while INN is
receiving news. At all.
No matter what the books tell you, you MUST run INN. It has my highest
recommendation.
+.# Let's back up for a minute. What IS news, anyway?
News is many things to many people. To start with, it's not really
news at all. The closest analogy is a gigantic distributed public
message system with its own customs, folkways and lore. Generally,
the more intelligent Internet users read and write news, leaving
Internet Relay Chat (IRC) to the dull (see below).
News is available on over 7,800 different topics, which represent
just about everything there is in our universe - and even in many
universes yet to be discovered.
Just to complicate things and make them interesting to the unwary,
News is not Internet News; it's USENET News. What this means is that
more primitive systems with only UUCP (Unix to Unix Copy) access can
participate in news. Other Internet services, with the notable
exception of electronic mail, require a direct Internet connection.
+.# Which mail transport agent should I run?
Most likely sendmail, despite its fearsome reputation. If you arm
yourself with The Bat, aka the book "Sendmail" by Bryan Costales, Eric
Allman and Neil Rickert, it's not too hard to modify an existing
sendmail configuration file for your site. The tutorial walks you
through all the various commands and the steps you have to take to
customize it as needed. There's surprisingly little you really need
to do or understand to set up a basic file.
Eric S Raymond <esr@snark.thrysus.com> was kind enough to give me a
sizable amount of information about sendmail and its chief rival,
smail. Sendmail presently wins over its rival by allowing multiple
transactions (that is, posts of mail) per SMTP connect; smail is not
so generous. This means massively poorer efficiency for smail and
thus his present use of sendmail. Eric says: "This makes them
significantly less efficient than sendmail for sites with high mail
volume. When this changes, CCIL [Eric's system] will move to Smail
3.0 and drop sendmail *instantly*."
As a BSD user, Eric has this to say about the task of sendmail
installation: "The stock sendmail V8 supplied with BSD/386 1.1 and up
works pretty well. The V8 developers did a fair job of hiding the
mind-wrenching ugliness of sendmail config files behind a layer of
more civilized m4 macros. Note: be sure you install the CERT security
patches for V8, or that your vendor has done so."
Tony Sanders <sanders@bsdi.com> adds: "You should mention that people
want to use sendmail V8.6.9 (or better) and that it's available from
ftp.cs.berkekey.edu:ucb/sendmail." This includes the CERT security
patches, so you don't need to worry about them.
As a Linux user, here's my experience: I used sendmail IDA, which is
available (from sunsite.unc.edu) pre-configured for Linux users.
This, naturally, makes IDA the logical release for that system. I
attempted to compile Berkley Sendmail V8 on my Linux box, but found it
impossible without the installation of massive amounts of additional
software. As a result, I decided to give up on V8 and use IDA. I
found The Bat to be a faithful friend once I got to know it. After
reading appropriate parts of The Bat, I was even able to write my own
"Mail Delivery Agent" to convert standard network mail into
BBS format mail (stored on my system as consecutive text files, just
like news). It was easy and even fun; my opinion of Sendmail
improved dramatically when I read the parts of The Bat that
effectively took me by the hand and told me how that could work.
Both Eric and David W Rankin Jr <rankin@ms.uky.edu> have about the
same opinion of smail. They tell me that it's relatively easy to
configure, particulary for mailing lists. Unfortunately, it's
hampered, as said earlier, by mediocre SMTP support. David Rankin
also notes that it's not well suited to complex hub/spoke routing
schemes. As a result, you might well wind up getting smail while
it's suitable for your small site, and then having to configure
sendmail or another more powerful program later.
David W Rankin Jr <rankin@ms.uky.edu> also puts in his recommendation of
MMDF, an older mail transport agent which is now not too well known.
"In case you're not familiar with it, MMDF is a mail transport system
(like sendmail or smail) designed for higher traffic sites that do
multiple protocols or lots of routing. MMDF is older than sendmail,
generally more secure than sendmail (it uses several programs,
isolating the root parts. Of course MMDF has been pounded less than
sendmail), and a LOT easier to configure (IMHO)." His present site
uses MMDF, as will any sites he administrates. However, he has not
yet configured MMDF; he has only seen it successfully in action. He
says, "take all I say about it with a shaker of salt :)". MMDF
can be obtained via FTP from a.cs.okstate.edu.
+.# How do I set up a Gopher server?
The Gopher server is proprietary property of the University of Minnesota
and is available to the commercial world only under rather stiff terms.
"Commercial use" requires a $ 500/year payment, regardless of the size
of your site. If you use it in a direct order taking application, they
want a percentage of profits.
Gnu is apparently coming out with a free gopher, but aside from that,
I have no details. Tony Sanders <sanders@bsdi.com> suggests that
you check into "gn", the new Gnu Gopher. It's available from:
ftp://ftp.acns.nwu.edu/pub/gn/gn-1.0.tar.Z
The newsgroup comp.infosystems.gopher covers gopher more fully. For
the most part, World Wide Web (WWW) seems to be replacing Gopher as
the information distribution tool of choice.
Gopher clients seem especially vunerable to running away when
the user hangs up on them, using incredible gobs of CPU time.
This is also true of Internet Relay Chat (IRC); see the IRC
section. I have included my perl script for handling this problem in
the technical problems section.
Gopher does have one significant advantage over WWW, however: its
more formally organized nature makes it easier for new users to
understand.
+.# What about the World Wide Web?
WWW servers appear to be free; I have not yet set one up, but plan to.
Depending on the nature of your users and what you allow them to
offer in their personal WWW pages, your server may be subject to some
truly awesome loads. Be sure to take this into account when
determining fees for these services. A reasonable compromise might
be to allow people to put their data on the Web at no charge as long
as they used no pictures, and charge for any pictures presented in
some way. The newsgroup comp.infosystems.www covers the World Wide
Web.
For more complete information on the World Wide Web, Tony Sanders
<sanders@bsdi.com> suggests the following URL:
http://www.bsdi.com/server/doc/web-info.html
Mosaic is the most well known WWW application; it is graphically-
oriented and works only on direct or SLIP/PPP connections. It's
probably the main reason many of your users will demand SLIP. If you
can't presently offer SLIP, you might tell your users that people who
do use Mosaic tend to condemn its snazzy graphics as too slow to be
practical. Mosaic is perhaps a tool ahead of its time, at least in
fully graphical mode; we can look forward to the eventual time when
people will have 56k links and can then take advantage of its full
power. Fortunately, you can compromise and disable graphics to
regain usability.
If you are interested in offering World Wide Web server services, you
should be on the inet-marketing list. This list includes many
fascinating discussions of marketing via the Web, including the
proper pricing of a Web presence.
Lynx, a VT-100 oriented WWW client, works very well if you don't
absolutely need graphics. The graphics can be automatically
downloaded to the client system if desired. It's worth noting that
many Web users disable the graphics because they come up very slowly
on a typical 14.4kbps SLIP connection. I have now used Lynx to download
graphics I wanted to see, and it really does work quite well.
+.# What about Internet Relay Chat (IRC)?
Internet Relay Chat is probably the most brain-dead use of the
Internet short of downloading X-rated pictures from the binaries
groups. Of all the things you will offer on your system, it's
probably the best way to attract clueless but paying customers.
Personally, I am vaguely nauseated whenever I engage in an IRC
conversation; the atmosphere is dismally tacky, the people
uninteresting, the conversations hideously dull and the software
dreadful. (If IRC fans in the audience want to point out places
where this is not true, they should feel free to drop me a line).
Despite all this - and probably because of it - the average user
loves IRC, and it probably puts fewer demands on your system
than just about any other service. As I would say, you don't
have to like it to offer it.
To add a little fairness to the above, I will say that IRC
is no worse than any other multi-user chat system I've seen,
such as those offered by various TBBS and Major BBS bulletin
board systems. Still, commands such as "kick" and fearsome
creatures called "bots" do not make for a pleasantly civilized
atmosphere.
IRC works by connecting you to an IRC server. There are two IRC
networks: Eris-Free Net (EFNet) and Undernet. Because IRC is a
networked resource that uses the entire Internet for its
conversations, the two nets are the only points of difference; any
IRC server on Undernet will be identical to any other IRC server
there, and the same is true for EFNet. Choosing the closest server
will not deprive your users of anything; it will, however, decrease
loads on the Internet, and provide them with far superior
performance. Because of this, taking a few minutes to find that
server is strongly recommended before offering IRC to your users. I
would suggest offering two IRC commands - one to hook up to EFNet and
the other to attach to the Undernet. EFNet is the huge one; Undernet
seems to be run by significantly more clueful people, but there's
often nothing going on there.
IRC was designed to perform operations enjoyed by the lowest
common denominator, but since it is a typical Unix program
designed by university types, it has tons of confusing options.
I recommend that you study the Undernet's IRC FAQ, available
at [[ pointer to undernet faq ] and make it easily available
to IRC users. It should answer most of the questions that
come up. Please note that the Undernet FAQ actually covers
operation of the IRC system itself and not only Undernet.
Even if you don't want to connect to Undernet (it's much less
popular than the older Eris-Free net), you still want to
read the FAQ and make it available.
Fortunately, the Undernet people have also made it very easy to
set up an IRC client. They have prepared IRC clients for various
systems at their FTP site [[ find URL ]], and setup for them
is clearly explained and quite easy. I recommend this strongly
over trying to compile your own IRC client; I attempted that
initially and ran into errors. The pre-compiled binaries from
Undernet work just fine.
As mentioned previously, IRC client software is not terribly well
written and will very often "run away" from your users. This will
cause it to fail completely, grinding up amazing amounts of CPU time
in the process. See "Dealing with Runaway processes", below.
+ What about Fees, Terms and Conditions?
I am shortly going to be writing a separate document on condition-
related issues; hopefully I will be able to write a suggested
terms and conditions list for all providers.
+.# How much can I charge? How much do other providers charge?
The largest provider, Netcom, charges $ 19.95 per month, or $ 17.95 if you
bill off a credit card. This is for full, unlimited access. Before they
pulled it for repairs, they were charging $ 17.95 per month plus $ 2/hour
for SLIP/PPP. Their new NetCruiser service, which is effectively SLIP/
PPP in disguise, costs $ 19.95 per month, including 40 free hours. They
also offer unlimited free access during "off-peak" hours, which are
defined rather oddly: Midnight to 9:00am. For hours past the initial
40, the rate is $ 2/hour.
Netcom has POPs in many major cities, so statistically it's the most likely
competitor for you to come up against. CRL, a major competitor, charges
almost exactly the same rates.
Netcom gained a richly-deserved reputation for dismally bad service
over the last several months. For a while, they seemed to have cleaned up
their act, but it's now as dreadful as ever. Their NetCruiser
service was also panned recently by critics as too slow and poorly
designed.
IBM's new network provider service, which will sell SLIP accounts
designed to operate with its new OS/2 Warp Internet Connectivity kit,
is charging $ 12.95/month for up to 6 hours of use, $ 30/month for
up to 30 hours. These rates are very competitive and are bound to
make the Internet access industry a great deal more competitive in
the future.
Karl Denninger's MCSNet charges $ 60/3 months, or $ 240/year. If you stay
with them for a year, you get two free months, reducing your effective
cost to $ 17/month. For SLIP/PPP, the 14-month cost is $ 260, under the
same terms.
+.# How can I distinguish my service from the competition?
In the beginning, low rates, friendliness of service and lack of heavy
system load may be your best bet, at least when competing with the
national providers.
I'm using unique software I wrote myself as a lure, but I recognize that
this is not an option for many people. (See the section on BBS software,
below).
+.# Accounting and Billing Practices (*)
Billing by credit card has major advantages for both you and your
customer. It's definitely the best option for most providers, but it
may be exceedingly difficult to get. See below.
There are a number of services now that will set up 900 number
billing for you. Your customer calls the number and gets a code; he
then calls your system and types the code in. The system is then
unlocked for a specific period. You are paid at the end of the telco
billing cycle. This is probably one of the easiest options to
qualify for, but you only get 2/3 of the total charge; this might be
unacceptable to you. People who have used the 900 number and
received your services may also attempt to contest the charges on
their bill. Due to the sleazy reputation of 900 number vendors and
systems, this is normally granted as a matter of course by the local
telco.
Finally, you can get people to pay by check. You get 100% of the charge,
and you get it now. The bad news is that many people who will blithely
charge billions on a credit card have an amazing resistance to getting
around to sending a check in the mail. Unless you have your strongman
Guido around to collect from your customers, this might not be the best
solution. You should always accept this form of payment, since it's
the simplest type for you, but you should not count on your customers
doing it. It's generally best to offer tempting discounts for longer
terms, since it's a big hassle getting people to send in a check every
month.
+.# Credit Cards (*)
There are two issues surrounding the world of credit cards. The first
is how big a percentage they nick you, and the second is whether you
can get them to accept your company at all.
Most credit card processing companies are leery of any businesses selling
services, especially out of the home. You will find it much easier to
get their attention if you have physical business premises, instead of
your garage.
A company called Teleflora - that's right, a florist delivery service -
has made quite a nice little sideline business out of processing credit
card orders for bulletin boards and similar operations. However, they
now appear to be out of this business (as of November 7, 1994).
Chris Hawkinson <chris@mhv.net> notes two interesting things. First,
if you charge by the hour, you are counted as a worse risk than if you
charge a flat-rate fee. To accept VISA, you must do the following if
you have an hourly rate plan:
* Pay a special annual fee
* Have the worst discount rate
* Have an 800 number which appears on credit card statements.
Second, just because you have received money, it doesn't mean you won't
have to give it back. Chargebacks can occur up to a year later with
just cause. "Further, since an online system usually doesn't have the
normal signature receipt, the customer is almost ALWAYS right."
Finally, he notes that extremely tight security is necessary for a
system that keeps credit cards information online, since crackers are
attracted to card numbers like ants are to sugar. Perhaps the best
way to do this is take the card numbers and copy them to a remote
system inaccesible to the public.
Chris Myer (cmyer@su102a.ess.harris.com) did a complete survey
of net.wisdom on this topic. He came up with the following
additional comments on how you might get accepted for a merchant
account, and what kinds of rates:
1 Get an account through a local bank. Chris says this isn't very
likely, unless you have very substantial dealings with them. "These
guys don't understand the 'net, and will simply consider you a
mail-order business. If you are home-based, forget it. Your only
hope is to get an employee of the local bank to go to bat for you--in
a big way!" Karl Denninger <karl@mcs.com> apparently got his own
merchant account in this manner; keep in mind, however, that his
five-digit account balances probably helped just a little.
2 Independent Service Organization. These people serve as
go-betweens, insulating you from the more stringent standards
banks require. According to Chris, "Many of these companies
charge horrendous up-front, non-refundable fees, but some are
reasonable."
The most well-known of them is Teleflora. Evidently these guys
decided that they would become ISO's, and they are big believers
in BBS's. Fairly large non-refundable application fee.
Unfortunately, they seem to be leaving the business at this
time. (800) 325-4849. (x2076?)
Card Service is another outfit that's been mentioned. Unfortunately,
little information is available about their serivces. Card Service,
(800) 944-7164, or try (800) 947-3650 (faxback, request 765.)
3 Discover Card: "Evidently, they not only provide Discover
cards, but once you get one of those they will also provide access
to Visa/MC merchant accounts. Call (800) DISCOVER (347-2683 for
the alpha-numerically challenged) and ask for Merchant Services."
Here's the information from Chris on what kind of rates you can
expect to pay:
Samples of Rates Charged:
=========================
Visa/MC: 2.20%
2.25% plus $.25 per transaction, plus $2.00 statement
charge.
2.32% plus $5.00 per month.
3.0%
3.5% plus $.30 per transaction.
4.0%
Discover: 1.9%
3.25%
AMEX: 3.25%
4.5%
Terminal: $21 per month, or $200 outright.
$600 outright, $75 for repair.
The following is copied straight from Chris' message:
Suggestions for Getting Accepted:
1 Try to avoid being labeled "home-business", "mail order" or
"telemarketer". I don't recommend lying if you fall squarely into
one of these catagories, but don't let them put you there if
you're not.
2 Minimize the amount you predict you will charge. Due to the
fact that charge-backs can be made for up to 6 months after a
transaction, the bank's amount "at risk" is 6 times the predicted
monthly amount.
3 When working with a bank be prepared to keep liquid assets in
the bank equal to or greater than their "at risk" amount. This is
usually necessary for no more than 6-12 months of good charges.
Suggestions for Protecting Yourself:
1 Try to find some way to get a signature on file. One way I
thought of was not only to have them fax in the card number and
their signature, but a photocopy of their picture ID with signa-
ture on it.
2 Make sure you have a way to verify that the ship-to address
for your product (if you are a retailer) is the same as the
billing address of the credit card holder.
+.# Making the Internet more User-Friendly
Several different approaches have been taken to increase the user
friendliness of the Internet, traditionally a rather cold place
for new users.
+.# What are SLIP and PPP?
SLIP and PPP are serial protocols that get your customers directly on
the Internet, so they can use tools on their own computers. This has
three basic benefits: (1) they can get easier to use Windows or
Macintosh versions of Internet software, and control exactly which
programs they use; (2) they can receive graphics directly over the
net; and (3) they don't use power from your CPU unless they are
accessing your disk (by reading news, for example). Unfortunately,
it is quite complex to set up, although most people figure it out
eventually.
SLIP and PPP appear to be the future of the Internet. There is an
increasing amount of software available that will only run in this
way, and most of it is in the graphical format appealing to consumers.
Most people who run an ISP and want to be able to charge significant
sums of money for access need to offer SLIP/PPP.
You need something called a terminal server to offer SLIP. It's
very expensive. (See the hardware sections).
Some programs have now been written that simulate SLIP for shell
account users. They will work, but are harder on your system load
(because they use your shell account computers) than real SLIP/PPP.
This should significantly lower the cost of SLIP accounts in the
future, a trend that has already developed considerable steam.
One of these programs is called The Internet Adapter (TIA); some
kind soul should give me a pointer to it. I believe a single-
user license is $ 25, and a full system site license is $ 495.
A free trial period is available.
IBM already has SLIP client software in their new OS/2 Warp product.
Microsoft is planning to add this to Windows95 as well. So it should
become easier to operate SLIP services as time goes on.
Note that the CIX (see our section on them) requires membership if
you are relying on a CIX provider for your connection and you resell
SLIP/PPP. How this applies to TIA (the simulated SLIP/PPP) is
unclear at this time.
+.# What about proprietary GUIs, such as the Pipeline?
The Pipeline, a NYC-based Internet provider, took a different approach.
They wrote their own customized terminal program that basically makes
their system look like SLIP/PPP without all the headaches. It's clever
and works well. Unfortunately, as of last writing, they were asking
a rather massive price for it. You can try it free of charge on their
demo line; telnet to pipeline.com.
I am planning to write a similar GUI front end for my own customized
software, but it will probably be a while before I have it done.
+.# What about BBS Interfaces?
Bulletin board systems, such as TBBS and Major BBS, are rushing into
the Internet bandwagon. Traditionally, the interfaces used by both
these programs are far too clunky for efficient newsreading, and
that is virtually the only service they've been able to offer. How
much this changes will determine how much influence BBS vendors finally
have on the Internet world.
I am presently writing a customized BBS interface that I rather like.
I'd better like it, since I wrote it. :-) Telnet to amazing.cinenet.net
or call (818) 997-7500 to check it out at no charge.
Eric S Raymond <esr@snark.thyrsus.com> has been working on some
similar projects. Like many projects of its kind, this was inspired
by dissatisfaction with existing systems:
"I know of three major projects to produce a UNIX BBS interface suitable
for use by ISPs -- FreePort, InterLink, and Chebucto Suite.
"FreePort is the BBS interface used by NPTN, the National Public
Telecomputing Network, and its affiliated Free-Nets. Despite its
name, it costs $850 to license from Case Western Reserve University.
And it is an utterly awful implementation of a clumsy, stupid design.
Most of the traffic on its lists, freenet-tech and freenet-admin, is
disgruntled Free-Netters wishing desperately they had something
better. To see for yourself, rlogin as `visitor' at freenet.fsu.edu
or yfn2.ysu.edu."
I - the FAQ maintainer - wanted to check out this claim, so I did
exactly as Eric suggested. Sure enough, the Free-Net software is
awfully clunky for an $ 895 program. It's inarguably simple, but
the user interface is far from the best. It looks a lot like a
stripped down version of Galacticomm's Major BBS; unfortunately,
that's not a compliment.
Eric again: "InterLink is the interface I wrote out of complete disgust
with FreePort. It combines the ease of use of a menu-based system
with the feel of a MUD (multi-user dungeon) game; it has unique
features for decentralized administration; and the code is simple,
flexible, and easily extensible. This software is production-quality,
supporting more than 800 users as of August 1994, and is far superior
to FreePort in every respect, but I have not yet released it; I'm in
the midst of a redesign for better integration with WWW. To check it
out, telnet to locke.ccil.org."
I checked it out, and it is indeed a slick, nicely designed and
easy to understand program. It has a much more modern look and
feel than the FreeNet software, and it's considerably easier to
figure out.
The David's Amazing Internet Services software - otherwise known
as my own - is very different from either program. Anyone interested
in checking out the state of the Internet BBS art should definitely
take a look at all three systems. Which one is better? I'd say
it depends on your own personal taste.
Eric continues:
"There is a project called `Chebucto Suite' under development in
Canada, with goals similar to those of InterLink. From their
prospectus, it appears that they already have full WWW integration
(the Chebucto browser is a souped-up version of the WWW lynx client).
They appear to be behind InterLink in some other areas, notably
administration tools. I don't know where Chebucto Suite is actually
running; contact David Trueman <david@cs.dal.ca> for information.
"It seems likely that Chebucto Suite and InterLink will converge over
time, as both teams have expressed interest in technology exchange with each
other. Both seem likely to be available for ISP use in early 1995."
The FAQ maintainer has now written to David Trueman, and will try to
review his system as well.
+.# Is there an easy, painless way to provide Internet services?
Possibly. Draper Kauffman <draperk@io.com> has a beguilingly simple
business idea to share with potential providers in underserved regions
of the world: sell telnet-only accounts to his system. This way, he
notes, you lose a lot of headaches. No need for gigabyte drives;
use his. No need to offer shell accounts and stay up all night worrying
about infamous security holes; he'll offer them for you. No need to
contract with credit card companies; he knows how tiresome that is to
new businesses and takes all of it out of your hands. You charge what
you want and it all goes through his accounting system. Anything over
a $ 15/month flat fee is given to you as a nice convenient check you
can deposit right into your account.
The main advantage of this idea is that you have 24 hour a day 7 day
a week monitoring and support by his already-existant ops center. This
would be horribly expensive for you to do yourself.
The main disadvantage of this idea is that you have no control over
your own system; in reality, you're selling someone else's product and
not your own. If you (like me) are starting a provider to put your
own personal stamp on the world, this is definitely not for you.
However, if you want money and you're in a market with low
competition, you could probably charge $ 25/month and pocket the $ 10
difference without doing anything more strenuous than maintaining
your modems and router.
Note that, since this would be a telnet only situation, you would
almost certainly require a T1 connection, since 100% of your users'
time would be using the Internet itself quite intensely. The high
bandwidth required might offset some of the administrative savings
of the proposed plan.
However, it can be an effective way to build a strong customer base
at the start, where cash for hardware, software and customer support
staff is hard to come by. You also have more time for marketing your
service, and it's easier to sell a full-service account for $ 25
than a semi-reliable account on a start-up system for $ 18.
Draper suggests that your initial goal should be to build up your
system in stages to handle 750 to 1,000 customers using 64 lines and
modems and a T-1 line. If you charge $ 25 per customer and pay
about $ 30/line and $ 2,000 for your T-1, you net somewhere around
$ 4,500 to $ 6,000 per month with no payroll. Then, with this kind
of strong cash flow, you can easily add the drives, CPUs and staff
you need for complete services, and gradually bring your customers
back to your system full-time.
+ What sorts of technical problems should I expect?
Thanks to Tony Sanders <sanders@bsdi.com> for contributing this list
of some of the many problems Internet connectivity is heir to:
Internet connection line failures (dealing with telco and provider)
Routing problems
General network problems (site x.y.z is down and the user complains to you)
Catastrophic Disk failure (make sure you have backups)
Users deleting files and then wanting them back (backups again)
Modem problems all of sorts (parity mismatch, hung modems, flow control,
not disconnecting properly, not connecting properly)
System crashes and hangs
Configuring ftp, telnet, shell accounts, nntp, www, smtp/sendmail, nfs
Bug tracking (a potentially big problem)
Billing
+.# What can be done about System Crashes?
Ideally, there should be someone around 24 hours a day to make sure the
system is put back up when it dies. A reasonable compromise for companies
that aren't large enough to do this is to be sure someone's on the system
during peak usage hours, to make sure the system is reset when it goes down.
(Information on devices to reset the system automatically upon a crash
should be given here).
You may be able to prevent many system crashes - or at least minimize
their effect - by purchasing an Uninterruptable Power Supply (UPS) and
a mechanism to automatically shut the system down gracefully upon power
failure. Unfortunately, this is another issue I'm not that familiar with;
expertise from those who know would be appreciated.
During my current business trip, when I've had to be away from the system
for almost two weeks, I asked my roommate's girlfriend to check the
system and reset it when "nothing appears when you hit the <enter>
key on the console." Apparently, she's already done this once, and
it's worked fine. If you can explain the care of your system easily
enough, just about anyone who likes you should be able to help.
(I wouldn't recommend handing this off to an enemy, since it's all
too easy to destroy your system this way! :-( )
+.# What can be done about Network Outages?
When part of your network link fails, you're dead in the water.
If you're using a SLIP connection for your network link, you will find
yourself disconnected occasionally. I have written a program to automatically
reconnect myself when this happens. It runs every 15 minutes (through
cron) and checks to see if there is a DIP process running. If there is
no DIP running, it starts one up. This lets my system automatically reconnect
even when I'm not there to tend it. Here's the program:
/* dipcheck.c -- check to see if DIP is running
By David H Dennis * david@amazing.cinenet.net
This program is hereby placed in the public domain; no warranty
exists, expressed or implied.
*/
#include <stdio.h>
#include <unistd.h>
#define FN "/tmp/dipps"
#define LEN 100
main()
{
char s[LEN];
int ct = 0;
FILE *fp;
system("ps -aux | grep dip >/tmp/dipps");
fp = fopen(FN, "r");
while (fgets(s, LEN, fp)) {
ct++;
}
fclose(fp);
/* The PS and DIPCHECK commands also contain the word 'dip', so
if there are less than 3 uses of the word found by grep, we're not
connected, and an attempt should be made to reattach ourselves */
if (ct < 3) {
printf("Executing DIP ...\n");
system("/user/dip/dip /user/dip/sample.dip >/tmp/dipout");
/* Note: Hollywood is the name of my network connection */
system("route add hollywood");
system("route add default gw hollywood");
}
}
Someone will probably flame me for writing this in C, when it would
have been more elegant as perl or even a shell script, but who has
time to learn them? :-(
According to Tony Sanders <sanders@bsdi.com>, "Perl is a godsend for
ISP's willing to learn it. It is the swiss army knife of the system
administrator, a real survival tool." I finally broke down and learned
it to write the "kill runaway processes" program, which will be presented
later.
If a 56k or T1 connect fails, Tony Sanders <sanders@bsdi.com> says
that you first reset everything on your end. The CSU/DSU (at least
the ones I've seen, cheap models might not have it) has various local
and remote loopback tests it can perform, do those. If those fail
you call your provider and/or the telco.
+.# Hung Modems
Rackmount modems really score here; as mentioned previously, there are
complex diagnostics and re-routing systems built into the modems. If,
as seems more likely, you have a bunch of tangled wires leading to
heaps of external modems, you will have to find out which one is causing
the problem and reset it. This can usually be done easily enough by
switching it off and on.
If you are calling the system from a remote site, and find it rings and
rings thanks to a bum modem, you can transfer to the next line by calling
the main number on your voice line, and then calling the same number on
your data line. You should then get the next line on the rotary, which
is (hopefully) active. Then, it's a fairly simple matter to inspect ps,
find the runaway job, and kill it. Usually that will reset the modem,
and the system will once again work.
+.# What would be a good backup policy
Off the top of my head (another section that needs to be fleshed out
with some real-world opinions), you should back up your system and
user files daily, probably with a seven-day rotating backup
procedure. I wouldn't be worried about news; lost news tends to be a
self-correcting problem.
Recommendations on backup equipment and procedures would be appreciated
here.
+.# What services are particulary hard on performance?
What should I do about them?
My thanks to Alicia Salomon <salomon@seas.gwu.edu> for reminding me to
ask this question, and supplying part of the answer.
Most providers will start with a single computer performing all functions,
including mail, news, ftp, www serving and user processing.
Because news flows into the system constantly, and since its processing
can put a significant burden on the system's disks, this is normally
one of the first functions to be transferred to a separate machine. Since
the advent of INN, this is not nearly as much of a problem as it once
was, but this is still sound advice.
Tony Sanders <sanders@bsdi.com> notes that news is *very* hard on
your network link unless you have T1. It's about 25% of a 56K line.
Tony recommends PageSat, as shown elsewhere in the FAQ. I know of
at least one person who got a PageSat connection and is highly
dissatisfied with it, so I'd like to hear from other PageSat users,
pro and con.
Other processes often put on dedicated machines include FTP, Gopher
and WWW. FTP in particular can put an enormous strain on system
disks, especially if users are allowed to place popular files in their
own directories. There was an enormous stink created on Netcom when
it was discovered that some users' FTP directories had X-rated
pictures in them, and that they constituted some 60% of the total
bytes downloaded from all of Netcom, causing vast overloads on Netcom's
machines. It might be a very good idea to devote a machine with a
large local disk to the user directories and transfer all the load off
the main system.
Potential load from Gopher and WWW could be immense, particulary
if massive image files are involved. It would probably be a good
idea to use the FTP machine for user Web and Gopher pages as well.
Tony Sanders notes the following: "Well, the real point of load comes
from how popular the information is. The servers of the
Shoemaker-Levey comet photos got creamed as thousands and thousands
of people requested the pictures. The good news is that you can
charge some serious money to local business to put up information on
WWW. That alone could probably pay for a T1 line."
Multi-User Dungeons, or MUDs, are "virtual world" games that account
for a large percentage of the Internet's popularity. Karl Denninger
<Karl@MCS.COM> allows them on his system because he had a spare SPARC
10 sitting around with nothing better to do. If you are not so lucky,
MUDs could devour your system memory and effectively kill performance
totally. Here is another case where a dedicated machine would seem
almost a necessity.
Note that telnetting TO MUDs is a very easy thing to do; operating
one on your system is the complex and compute-intensive burden Karl's
taking about here.
+.# What can be done about users who walk away from the keyboard?
Write or obtain an idle timeout program. Usually the archives for
your operating system will have something that will do. For Linux,
ftp to Sunsite.unc.edu and get /pub/Linux/system/Admin/idleout.tar.Z.
There is a certain degree of controversy over how long the idle timeout
should be. Netcom uses 10 minutes, which many people find too short.
MCS uses 20 minutes, which is probably about right.
I think it would be a good idea to vary the idle timeout depending on
the number of lines in use. During an extremely light load time, it
might be ok to make it as much as an hour. This can help users who
have to go to the bathroom or who got engaged in a long conversation,
and it doesn't much hurt the system. However, I have not yet
experimented with the idle timeout software.
Tony Sanders <sanders@bsdi.com> agrees with this idea, and suggests
that you might want to involve users in the process. Explain to them
why an idle timeout is necessary, and get their input. It is, after
all, trivial to defeat; just write a program that automatically sends
a space or two to hold the line. Another way to handle the idle
question is to disable the idle timer until nearly all the modem
lines are in use. Always remember to treat your users with courtesy.
"If you explain things to them the most of time they will go out of
their way to make things right."
+.# What can be done about users who never log out?
What about users who seem to be on the system 24 hours a day, 7 days
a week? This leads us to the controversial question of pricing.
Historically, Internet service providers have charged a fixed fee
per month, regardless of the amount of use made on the system. This
has almost always been the best model for customers; all but the
lightest users pay less than they would under the non-fixed schemes.
Unless they have unique offerings like the slick graphical interface
of NYC's Pipeline, new providers are not going to be able to come in
at higher fees than (say) Netcom or CRL. This pretty much eliminates
the idea of hourly fees for most.
Well, maybe. Draper Kauffman (draperk@io.com) notes that Netcom is in
his area at their normal rates. Despite having higher rates, his
system and other local competitors have not suffered; the reason, of
course, is Netcom's infamously terrible service. He suggests that
excellent service can still get $ 25/month or more.
Hourly fees are mandatory, of course, if you offer continental
US toll-free access. This can often be arranged at very competitive
hourly cost as compared to a toll call to your site.
Some services, particulary bulletin boards, undercut the typical
ISP monthly rate but restrict access to a certain amount of time
each day. The Pipeline offers a set number of "free" hours and
charges for any longer period of time spent online.
+.# Trouble Ticket Systems (*)
This is an issue for providers that presently have employees, so
I will let someone answer this who is in a better financial condition
than me. :-)
+ Who needs and wants Internet Services? How can we reach them?
This is definitely an incomplete and imperfect section; additions are
especially appreciated here.
+.# Graduating College Students
Students graduating from college have already gotten addicted to the net
and will pay reasonable money to gain it through a commercial provider.
Reach them through advertisements in alt.internet.access.wanted and
similar newsgroups. Postings on free college bulletin boards are also
likely targets.
+.# People with Internet access at work
There are many people who would like to use the Internet outside of
work. They will pay for an account that gives them privacy from their
employers, allowing them to participate in the "forbbidden" sections
of the net without fear of sanctions.
The internet access wanted newsgroups are probably the best ways to
find these people.
+.# Current callers to computer bulletin boards (BBSs)
Bulletin boards are getting hooked up to the Internet in massive numbers,
but the culture clash between them and people with "real" Internet access
is very strong. Most bulletin board systems don't offer newsreaders with
killfile or threading capabilities, and as a result the quality of access
tends to be very low. In addition, direct services like Gopher, FTP,
etc, tends to be either non-existant or available at substantial extra
cost. Although this situation is bound to change somewhat, my experience
is that BBS software vendors are remarkably clueless in this regard.
BBS callers are used to paying fairly minimal amounts of money for access,
but this seems to be changing; I know of several large systems that charge
around $ 15/month, which would get you a shell account from a number of
small providers.
The best way to compete with BBSs is to ask to be added to major BBS lists.
A good guest account or new user routine is a must; BBS callers are used
to getting a good taste of what they're getting before they have to pony
up the cash.
+.# The General Public
They are obviously intrigued, if you look at the sales figures of Internet
books and information. Certainly there has been a massive flood of Internet
stuff, such as the Time cover story, and that has no doubt piqued a
significant amount of interest.
However, it may be quite difficult to reach them. Flyers posted around
bookstores might be a good approach, since people who find Internet books
are undeniably interested in access. I'd love a few more ideas to add to
this section of the FAQ. :-)
+.# Some low-cost marketing ideas (*)
Draper Kauffman <draperk@io.com> suggests that we contact our local public
television station and donate 10-odd subscriptions for auction during pledge
week. "They were auctioned off two per night for five nights, got a lot
of comment because of the novelty, and went for good prices (i.e. more than
we charge). They also generated 30 to 40 paid subscriptions and excellent
name recognition in a good demographiccross-section of markets, including
some groups we would normally have trouble reaching.
"On a guess, if you cound referrals from these new customers, we'll probably
get 60 to 100 paid subscribers for the cost of supporting 10 customers
for a year. Most important for a startup: there's no actual cash outlay!
Compare that to thecost of producing a 60-second TV spot and having it
aired 10 times during prime time."
He adds that this gambit may be effective exactly once, since his
competitors are also going to be doing it next year, and the novelty
value will be gone. But for the virtually zero cost, it's a
phenomenal return. I might add that public radio might do nearly as
well, and many markets have more than one public radio station. So
even if this has been done already by another provider, you may want
to see if there are stations or outlets they've forgotten.
Alicia Salomon <salomon@seas.gwu.edu> recommends the Guerilla
Marketing series of books (Guerilla Marketing, Guerilla Marketing
Attack, Guerilla Marketing Weapons, etc), which include a number of
great marketing tips and tricks. Many of them apply only to
storefront businesses, but a healthy percentage of them will work just
fine for providers. I have read the books and wholeheartedly endorse
her recommendation.
Chris Hawkinson <Chris@mhv.net> added two interesting ideas:
> Offer "a low level of free access" to local libraries and schools.
"The publicity and good will alone can make it worthwhile. Further,
since it isn't full access (let's say just to Gopher), users can
quickly get hooked. Lastly, you can use the libraries as an
'information provider' to add more value at the local level.
> Offer a reward for new users. Everyone who gets a new user to join
Chris' system gets $ 5 for each $ 55 (yearly) BBS account.
+ Now that I have users, how should I deal with them?
NOTE: I AM NOT THE WORLD'S GREATEST EXPERT ON THESE ISSUES OR ANYTHING.
IF YOU WANT TO STAY OUT OF TROUBLE, BE SURE TO ASK A LAWYER ABOUT ANY
DISPUTABLE PROVISIONS HERE.
+.# How should I introduce my users to the unique Internet culture?
This is really a matter of USENET, electronic mail and mailing lists,
since most other resources are purely read-only. I had some long
conversations with a number of people heavily involved in net
administration, and they asked me to convey the following to new users
of the Internet:
* Tell them not to post for about two weeks to a month. It takes a while
to understand what's going on in the unique Internet culture. Absorb
what's going on before beginning to speak.
* Read Frequently Asked Questions files (FAQs) before asking for help.
There's a wealth of information there.
* Read the messages on news.announce.newusers. They give you a lot of
great information on how to pose questions properly and how to get
the most out of your access.
It should also be emphasized that, although the guidelines on mailing
lists are very much dependent on the list itself, generally users
should conduct themselves therein by following the same basic rules as
on USENET. Because mailing lists generally attract many fewer posters
than newsgroups, and since the software used to read them is usually
more primitive, "flaming" and other anti-social practices can cause
considerably more harm to a list than a newsgroup.
+.# What sort of disclaimers do I need to protect myself?
[legal matter - our legal advisor should answer this]
+.# Can minors be allowed to use unrestricted accounts?
With all the spicy stuff on TV, in movies and in books, all of which can
be easily accessed by minors, it rather surprises me that there's so much
fuss about a few pictures. But there is.
The answer appears to be, "Only if they have a note from their parents."
But I'd like to hear all views on this issue.
Note that this is not confined to picture files; newsgroups like
alt.sex and alt.sex.stories also contain questionable material. Even
such a tame sounding group as news.answers is a time bomb; eventually
that alt.sex.bondage FAQ is going to find its way in there.
+.# How can I prevent minors from seeing the sex stuff?
You can certainly restrict a lot of it, but I doubt that you'll get it all.
In particular, news.answers contains some sexual material. It would be
possible to put together an automated kill-file mechanism that would put
any group with the name 'sex' in its title, and hide messages with the
name 'sex' in their subject lines.
My BBS-like software has an interface that gives another possible answer to
this question. With my system, you can effectively lock people in to a
specific list of newsgroups selected by the operator. Since there's no
shell account, there's no way to fool with the active file or read the
groups from the news spool.
A normal newsreader has so many security holes that it's not going to
work well.
I am very much afraid that protecting minors from sexual material on
the net is just as hard as protecting them from sexual material in
real life. That is to say, impossible. Even with my BBS-like
software, they could still telnet to a site that has those groups. It
is a great pity this opens up so many potential legal problems.
+.# How should I deal with Netiquitte violations?
There are as many different opinions on this issue as there are system
administrators. The general rule, however, is quite simple: a single
complaint can probably be safely ignored as a dispute between users
and not true wrongdoing on the part of your user. Multiple complaints
should be investigated, and proper warnings given. Only after a user
has received complaints about multiple incidents should his account
be in true danger.
As a general rule, your user is innocent until proven guilty. Many
obnoxious groups, such as the infamous alt.syntax.tactical crowd,
will complain to you about a perfectly innocuous user in an effort to
get her or him kicked off.
The law firm of Canter & Seigel was kind enough to tell us all about
their Immigration Law services a few months back. Many of us appreciated
their help so much that we wrote to thank them personally for the
information. :-) Unfortunately, so many people were so effusive in
their thanks that they filled up the entire mail disk of C&S' service
provider!
With the kind of blatant nastiness of C&S, there is really no excuse
not to drop them straight away. Indirect did basically all the right
things in their response to the C&S incident:
* Yanked the C&S account instantly.
* Posted messages to news.admin.misc and other appropriate groups telling
the world about their response. This helped stop the mail bombs and
other attempted punishments of Indirect.
* Set up a gopher server with fuller information about the incident.
(I welcome additional information on dealing with problem users and such)
+.# How should I deal with disputes between users?
Unless it looks like they are going to physical blows, you should
ignore most of these disputes. Dealing with physical threats is a
more serious matter. I have actually had that problem on a BBS I
operated, and my course was to ignore them. What I actually should
have done is open to debate, and I'd like to hear opinions from
others who have faced similar problems.
+ Legal Issues
THIS SECTION WAS NOT CREATED BY A LAWYER, AND THEREFORE DOES NOT CONTAIN
LEGAL ADVICE. YOU SHOULD CHECK WITH YOUR LAWYER BEFORE PROCEEDING.
INFORMATION CONTAINED IN THIS SECTION IS BASED ON MY REASONABLE KNOWLEDGE
AND BELIEF BUT IS NOT INTENDED TO BE COMPLETE.
That said, here are some of the legal issues currently facing providers:
COPYRIGHT: Many newsgroups contain blatant violations of copyright
law. The most prominent among these are the
alt.binaries.pictures.erotica* groups, which often include scans from
Playboy magazine and other copyrighted material. Karl Denninger
(Karl@MCS.COM) avoids potential liability by not carrying these groups
at all, and strongly urges other providers to do the same. Others
note that there are frequent equally blatant copyright violations on a
large percentage of groups, including the full text of newspaper and
magazine articles often typed into news.admin.misc. A CRL user
proposed to create a test case of this by bragging about his post of a
specific copyright violation to alt.binaries.pictures.erotica,
including his PGP signature for verification. As far as I know, no
action has been taken against him for this. Certainly, there has
been no recorded incident of any action taken against Internet providers
or their users for copyright violations against Playboy or anyone else.
Playboy did win a case against a bulletin board operator who scanned in
images from Playboy and sold them via subscription.
SLANDER, LIBEL ET AL: Someone more knowledgeable than I will have to
write this section.
THE COMMON CARRIER ISSUE: This one likewise.
+ Books and other Resources
Linux Documentation Project (LDP). These are a series of on-line
books which can be obtained via FTP and printed. An excellent
argument for getting a PostScript printer, since that's the
only format that really works well. You can also buy copies
of these books and other Linux information as the Linux Bible,
which is carried by some bookstores.
ORA (O'Reilly & Associates). A publishing company that puts out
a number of helpful books on network administration. They are
clear and well written, with plenty of clear illustrations and
helpful explanations. Unfortunately, since versions of Unix
differ a great deal, they may not be completely accurate for
your system. They are one excellent argument for using Sun
workstations or BSDI Unix, since their examples should conform
to thoe systems with almost complete precision.
USENET FAQs and newsgroups. This is probably the best source
of information on administrating a Unix or Linux site -
especially if you're using Linux. The traffic on the
Linux newsgroups has to be experienced to be believed, and
most Linuxers are eager to help. I'm not sure how good th
net support is for Sun or BSDI; I know BSDI's technical
support is known to be excellent.
+ Glossary of Common Terms and Acronyms
Note: The idea for a glossary was suggested by Tony Sanders
<sanders@bsdi.com>. At present, it's woefully incomplete and
additions will be warmly appreciated.
Bandwidth. (1) The maximum amount of information that can be
transmitted at any given time. A 56k leased line connection, for
example, has 56k of bandwidth. (2) An abstract unit of measure
of the load something puts on an Internet connection. "Multimedia
through the Internet uses too much bandwidth for a 56k connection."
Often used in Usenet to refer to stupid questions or wasteful flames,
i.e. "this is all a waste of bandwidth".
Client. A program that is run by users on their machine. It issues
requests to a server, which is generally located on another system.
For example, here's how a Gopher conversation might work:
Connection
Client Program Server Program Active
Send me your menu Here it is (sends menu) |
Receives menu from server |
Displays the menu on the screen
Lets client choose an option
Client chooses option
Client finds out what machine |
to connect to and asks server |
on that machine for the data |
Client receives data Server sends data |
Client displays data to user
As you can see from this rough sketch, this off-loads a great deal of
stuff from the server program, so that it can process client requests
more efficiently. This also makes the system appear very fast, since
requests to page through the document can be handled by the client
and run as fast as the client can. This is dramatically different
from connecting to the remote system and retrieving the file from the
remote system screen by screen; depending on the remote system's
response time, this could appear many times slower than the
client/server approach.
For example, the FTP client takes the user commands to transfer
files and passes them to th FTP server. The server then sends the
file to the client, which then passes it on to the user as
appropriate.
Note that in the case of Gopher and WWW, the actual connections are
very brief. In the example above, the connection is only active
when there is a "|" in the "Connection Active" column. This makes
use of the server much more efficient then under protocols such as
Telnet, where a connection is established for the length of the
session.
CSLIP (Compressed SLIP). SLIP with compression for a more efficient
connection. See SLIP.
Flame. An insulting message, normally with little real content.
A Flame War is a seemingly endless exchange of such messages.
FTP. File Transfer Protocol. This refers to a protocol describing
the way files can be transferred over a TCP/IP network, such as
the Internet. The program used to implement this protocol is also
called FTP. Normally, a FTP program is included with your basic
networking software, and little needs to be done to make it work
on your system. However, if you want to offer anonymous FTP,
which lets users on the Internet obtain files from your system,
you are urged to consult a book on such as the O'Reilly TCP book.
GOPHER. A relatively simple-minded menu interface to the net,
Gopher presents a list of menu options. Each option can refer
to a resource in a different location on the Internet. Gopher
is especially useful for new user information, since it's
so simple to use; the World Wide Web is superior in every
other way. Unfortunately, there is a rather expensive
royalty that you must pay in order to use Gopher for anything
commercial; because of this and the ability of WWW to do more
interesting things, Gopher is fading in popularity. Veronica
is a service that lets you search Gopher servers around the
world for various keywords; it's one of th better ways to find
something specific through the Internet.
HTML. (HyperText Markup Language). This is the scheme used to
design World Wide Web pages. Raw HTML is strikingly similar to
the text formatters of old, while lamentably incompatible with
any text formatter language I know of. Fortunately, there are
numerous tools that can help you write HTML with reasonable
efficiency; check out WWW page [[ www page where this stuff
is ]] To do well as an Internet provider, you'll probably need
to learn how to deal with HTML, since you'll almost certainly
want to design your own WWW page to promote your system.
HTTP. (HyperText Transfer Protocol). This is the protocol
used for information transmitted over the World Wide Web (WWW).
IRC (Internet Relay Chat). This is a direct interactive way
for people to hold conversations using the computer. The level of
conversation tends to be pretty low, and rather on the dull side, but
many users love it. See the IRC section, earlier in this FAQ.
LYNX. A non-graphical WWW client suitable for shell or BBS
account users. See WWW (World Wide Web).
MOSAIC. A World Wide Web graphical client program, usable only
under SLIP connections. See WWW.
News, aka NetNews, USENET News, Internet News, etc. This is a
messaging system that is one of the most famous and popular parts
of the net. See the detailed FAQ section on news for details.
PPP. Point-to-Point Protocol. A newer and supposedly better way
to connect your site to the Internet via a single serial line.
See SLIP.
RFC. Request for Comment. This is an informal system for proposing
Internet standards. The technical people who work on the Internet
upload RFCs to the NIC, where they are given a number and published.
Many of them are later adopted as Internet standards.
Server. A program running on a remote system that provides
information to a client. See Client for a detailed explanation.
SLIP. Serial In-Line Protocol. This is one of several ways to
attach a computer to the Internet via a simple (and cheap) modem
connection. See the earlier discussion on connecting your system to
the Internet for additional information.
TELNET. Telnet is a program that lets you remotely log in to
any other system on the Internet (assuming you have access). In
the olden days, university systems used to give guest accounts
on their mainfame computers so members of the public who could
afford a modem and computr could take a peek. Unfortunately,
due to the higher level of obnoxiousness of current Internet
denizens, this has become far less common. People can, however,
use Telnet to connect to various bulletin boards and special
interest systems hooked up to the Internet.
TCP/IP. Transmission Control Protocol/Internet Protocol. The
protocol used to send information through the Internet
WWW (World Wide Web). This is probably the best Internet browsing
system - certainly the most fun one to use. It's not as simple to
use as Gopher, but it provides many more interesting options,
including integrated graphics if you use the Mosaic browser. See
the detailed section on the Web earlier in this document.
+ Resources
+.# Internet Mailing Lists
inet-access@earth.com (Email inet-access-request@earth.com)
This mailing list is a way to exchange tips, information and questions
with other ISPs. This FAQ was compiled with assistance from inet-access
members.
small-providers@netheaven.com (Email small-providers-request@netheaven.com)
This list is a low-volume list that's meant to fulfull basically the
same function as inet-access.
com-priv@psi.com (Email com-priv-request@psi.com)
The name's supposed to stand for Commercialization and Privatization
of the Internet, but it's really the battle of the titans. Watch in
awe as the CIX self-destructs and hear paranoid (but often accurate!)
speculations about what the telephone companies will do next. "Just
because you're paranoid doesn't really mean they're not out to get
you." Monitor the paranoid world by subscribing to com-priv. Or
don't, and get a good night's sleep.
WARNING: This is an extremely high volume list. I have been told
it's virtually impossible to get off of once you're added.
inet-marketing@einet.net (Email inet-marketing-request@einet.net)
This moderated contains many interesting discussions of the best way
to market over the Internet, the design of World Wide Web sites, and
so on. You should definitely get on this list if you have an
interest in these topics.
+.# Internet Newsgroups
alt.bbs.internet Newsgroup
This newsgroup occasionally contains discussion of similar issues
to com-priv and inet-access, but it's mostly people advertising
their Internet sites or looking for same. This is a good way
to advertise and/or scope out the competition.
alt.internet.access.wanted Newsgroup
Pleas for Internet access. Once your provider is set up, you should
monitor this group to find people who might be interested in its
services.
alt.internet.services Newsgroup
Still more pleas for access and advertisements of access providers.
alt.culture.internet Newsgroup
Assorted discussions of Internet related issues. Alas, not as interesting
as it sounds.
alt.internet.media-coverage Newsgroup
Interesting discussions of media coverage of the Internet. Low-
traffic.
alt.internet.talk-radio Newsgroup
Internet talk radio and sound bytes.
+.# USENET Oriented News Groups
news.admin.*
alt.current-events.net-abuse
These newsgroups contain information about the current flood of
"spamming", e-mail abuse and other topics relating to net
administration.
news.answers
This is a repository for all FAQ (Frequently Asked Questions) lists.
A fascinating place to browse; be sure to make it easy for your
users to access it.
news.announce.newusers
This contains information all new users should read, especially if
you haven't produced your own versions of it personalized for your
system.
+.# Books
The O'Reilly and Associates "Zoo" - the books with the funny animal
engravings on the covers - are universally popular, and with good
reason. Some of the information, however, is specific to Unix
environments the authors are familiar with. "System Performance
Tuning", for example, goes into detail about BSD and System V
performance measuring utilities, but little seems applicable towards
Linux.
Expecially recommended among the "Zoo" is the book 'Sendmail' by Brian
Costales et al. Although the Unix-Haters Handbook gulps at its size
and its incredible level of detail (it explains concepts anyone
running a Unix system should already be familiar with), it's
definitely a first-rate introduction to an otherwise confusing beast.
The Unix-Haters' Handbook, by Simson Garfinkel, Daniel Weise and
Steven Strassman, is excellent comic relief for any Unix system
administrator. It tells us very clearly why we should have all
bought LISP machines. Or VMS. Or ITS. Or TOPS-20.
For Linuxers, the books produced by the Linux Documentation Project
are first-rate. Fortunately, O'Reilly is apparently going to publish
some of them, hopefully giving them a wider distribution. You
should definitely look for the Linux Network Administration Guide,
which was far more helpful than any other book I've seen in teaching
me how to set up my two-station network (Sun 3/60 to Linux PC) and,
later, my SLIP connection. If you're cheap and have a fast PostScript
printer, they are also available for anonymous FTP from sunsite.unc.edu.
Check out the directory (if I recall correctly) pub/Linux/docs/ldp.
I don't recommend trying to read the documents in TROFF or TeX format,
a true exercise for the machochist!
+ Conclusion
We're in an exciting world here, the emerging world of global
communications some call "Cyberspace". Those of us who work to shape
this world in our own image are called Internet Service Providers.
It's an awesome responsibility. Let's try to make it work.
I hope you have found the information in this FAQ to be helpful and
of interest. However, to make this document truly useful, I need
your contributions, which will be credited as requested.
David Dennis
david@amazing.cinenet.net
dhd@netcom.com
APPENDIX A: Simulating a Router
Kevin Smith <sonetman@netcom.com> was kind enough to forward a message
he saw on USENET about using BSDI as a router. I hope this will help
those who have this problem in the future.
This information is reprinted through kind permission of the author,
Michael Galassi <nerd@percy.rain.com>. I dropped him a follow-up line
asking permission to reprint, and asking for a price for the
RISCom/N1. He tells me it's "under $ 500 but I don't remember how
much." You can contact SDL Communications at (508) 238-4490 for more
information and current pricing.
Tony Sanders <sanders@bsdi.com> offers the following as an update:
"These days you will probably be getting a RISCom/N2 card from SDL, they
come in single- and dual-port versions (very nice for setting up a hub).
You'll need a driver from BSDI for it as it has a different interface
from the N1 card (just ask support@bsdi.com for the N2 driver).
Everything else is pretty much the same."
This is the last item in the FAQ due to its incredible length (over
400 lines). If you are not interested in this specific subject, you
can stop reading now.
From: nerd@percy.rain.com (Michael Galassi)
Newsgroups: info.bsdi.users
Subject: RISCom/N1 summary of experience
Date: 18 Jul 94 22:13:34 GMT
Organization: University of Illinois at Urbana
Lines: 404
Approved: Usenet@ux1.cso.uiuc.edu
Message-ID: <m0qQ0wY-000CoyC@percy.rain.com>
Reply-To: nerd@percy.rain.com
NNTP-Posting-Host: ux1.cso.uiuc.edu
Originator: daemon@ux1.cso.uiuc.edu
Hi,
A week or so ago I requested tips & experience of the list to help me
in an impending conversion from a MorningStar Express to a BSDI box
containing a RISCom/N1. I've received several requests for this info
and two people have provided me with some help. Additionaly, when I
ran into hardware problems the people at BSDI were as usual quick to
respond, very competent, and nice to deal with (they didn't even sneer
at me for not recognizing my hardware problem for what it was). Great
bunch.
The end result is quite nice, performance is great, the interface to the
router is what I'm used to in my normal syadmin chores rather than some
vendor's idea of what is best, and the machine is truly flexible, I
want more ports I go to any PC store, buy the board, build the kernel
and I'm off and running.
For starters, the responses I received pointed me at a document available
on world.std.com (now ftp.std.com?), I failed to locate the doc there but
both people were kind enough to provide me with a copy of this document.
I've attached a copy of it at the end of this message.
First off, the pitfalls I ran into.
o My motherboard (noname vlb 486/33) which appeared to work nicely in
an enviroment including 3 ethernets and a DigiBoard routing SLIP did
not work well with the RISCom/N1, the symptoms were that within no
more than an hour of starting the rn0 interface would lock up, all
else seemed OK.
o The weather out there was nice and I would have rather have been on
my motorcycle.
o getting the IRQs, I/O addresses, and memory mappings right for
all the boards was hellish.
The solutions involved replacing the motherboard, buying a new bike
the next day as a consolation prize, and taking very carefull notes
*ON A MACHINE THAT DOES NOT RELY ON THE ROUTER*.
The basic steps involved:
Create a config file rn0 defined. CAREFULLY pick IRQs, memory & I/O
locations that don't clash and update your config file. At this point
you need to know if your IP provider supports CISCO's HDLC or PPP,
the later is more likely. Add a line that reads "options CISCO_HDLC"
or "options PPP" as needed. Build your new kernel and install it. At
this stage DON"T change any files in /etc such as netstart, rc.local,
hosts, or DNS configuration, you'll get that later.
Configure the switches on your N1 to match what you've set in the kernel
Shut down your machine and put the N1 in it.
Boot up ms-dog and run the diagnostics for the N1, the program is called
n1.exe and is in \dos on the floppy you got with your N1. There is
aparently more than one version of this program for mine did not match
the documentation that came with the N1. The program is close enough
to "self documenting" that you should be able to wing it. One pot-hole
I fell into, I configured my N1 for IRQ 10, same for my kernel, but,
the diagnostics won't let you use that interupt. Since I wanted my
board to run with the same config as the diagnostics had tested I
changed my config a bit. I would guess the longer you run the diagnostics
the better, I lost patience at about 30 minutes <blush>.
When you are bored with watching the diagnostics, power down and attach
the cable from the db25 on your N1 to whatever DSU you have, this should
cause some LED activity, diferent DSUs will work diferently. Punch reset
and watch the boot message carefully, you should see a line looking
somewhat like:
rn0 at isa0 iobase 0x220 irq 3 maddr 0xe0000-0xeffff
Obviously your numbers may be diferent from mine.
If you don't see this run strings -a /bsd | grep "rn%d", if you get no
output you messed up building the kernel. If you do get output cd to
your build directory and look at ioconf.c, make shure the numbers on the
line that ends in /* rn0 */ match what you gave in your configuration,
if they don't rerun config, redo your make depend & your make, reboot
and try again. If they do match they you have a hardware conflict or
an incorrectly configured board. Fix and repeat as needed.
Once you are up in multiuser mode log in as root, and manualy ifconfig
your board with the local and remote addresses, netmask and link flag,
I use:
ifconfig rn0 inet 199.2.108.234 199.2.108.233 netmask 255.255.255.240 link0
that is, the local side is at 234, the remote end is at 233, the netmask
is 0xfffffff0. The link0 means I run PPP on this link, its absence would
indicate CISCO's variant of HDLC framing.
You should now be able to ping the remote end of your link, in my case
"ping 199.2.108.233", if this is the case, you can add a line identical
to the one you typed above to your /etc/netstart and you are done!
This leavs a *small* matter of routing. If you are a leaf node you
can get away with "route add default <remove IP address>", otherwise
you will need to configure gated to do the "right thing", this is
left as an excercise for the reader.
Hope this has helped some, writing it down certainly helped me see
what I had done and realize what I'll do diferently next time.
If you run into problems doing any of this, drop me a line, I'll
be happy to review config files and other similar things to help
you out.
-michael
---cut here---
Using BSDI as an Internet Router
================================
This document describes the basic procedure for using the SDL N1 board in a
BSDI machine to implement Internet routing functionality. This includes useful
general information such as N1 setup which should be applicable in many
proprietary (non-Internet) network connections.
Introduction
============
The good news is that getting the N1 to work is almost as simple as plug'n
play. Additionally, my own tests and from talking to BSDI folks, confirm that
even a 20mhz 386 BSDI machine has more enough cpu power to move lotta packets;
I get better consistent thruput using BSDI/N1 than I did using a NAT router
(and no longer have to deal with a number of connection killing bugs that
plague the NAT router).
Note that typically when you dedicate a machine to something as important as
routing (and other site services such as mail forwarding, POP account, DNS,
proxy ftp) don't give people login shell accounts on that machine. Also, for
security reasons it might be a good idea to chmod 400 on the /dev/bpf* devices
(or disable access to those devices completely once you've debugged the setup).
The Environment
===============
The Internet connection is through a 56kbs leased line (PacBell ADN -
California) terminated using a Dowty DCP3080 CSU/DSU. The N1 board connects
to the CSU/DSU throught a V.35 interface using the cable supplied by SDL.
The host with the N1 board then gateways to other machines connected via
ethernet.
Installing The N1
=================
An overview of the steps involved:
o Determine available base I/O, interrupt, dual port ram of your
bsdi machine
o Figure out if you have to do anything special about caching
o Test the N1 to verify setup
o Build a new kernel
o Boot new kernel; basic N1 test
Setting base I/O, etc.
----------------------
The default BSDI N1 setup requires that the board's base I/O be set to 0x220,
uses interrupt 5, and assumes dual-port usage at 0xe0000; these are very
reasonable defaults but check your config to see if they will work for you.
Note that since the N1 is a 16-bit device, it requires a 128kb segment of
dual-port ram (see N1 manual).
To check for suitability of preceding defaults, execute the 'dmesg'
command to display your machines config. The dmesg command will list all
devices base I/O (iobase), interrupts (irq), and dual-port RAM use (maddr).
If there are no conflicts, set the boards S1 switch to use base I/O 0x0220 and
interrupt 5. The maddr range is set in the BSDI kernel config (as described
shortly). Install the board in selected machine.
Setting up for Caching
----------------------
Now determine if the machine with the N1 uses an external cache; if it does,
determince if it's a write-thru or write-back. Check your motherboard manual.
Higher quality and most newer motherboards use the write-back cache; with this
type of cache you do NOT have to configure the dual-port RAM area as non-cachable.
Older and cheaper mother boards use the write-thru cache; with this type of
cache you MUST configure the N1's dual port range as non-cachable. Depending on
other boards you have installed (ethernet), it is possible to run out of
non-cachable regions in which case you'll have to turn off the caching.
Some motherboards support both type of cache; determine which is enabled on
your board and act as needed.
Testing the N1
--------------
Boot the machine with DOS and run the N1.EXE test program (provided by SDL).
This program will verify basic functionality such as base I/O settings and
interrupt. The most important thing this program will do is verify that the
desirable dual port ram range works on your machine.
The default 0xe0000 address should work fine with most newer MBs using
write-back cache; however, older boards might have problems. For example,
the machine I installed my N1 in would only pass the dma test at address
range 0xa0000.
Address range 0xa0000 is also used by vga driver; since I don't run X/have
a vga card on this machine, when I rebuilt the kernel, I deleted the vga
driver. If you use an older board, be aware of special situations like this.
Bottom line is, the N1 must pass the dma test at the memory range you plan
on using. If it doesn't pass - don't go any further; things will not work.
Building the Kernel
-------------------
In configuring the kernel for the N1 you should disable any devices you don't
need since adding the rn0 device could result in a kernel that breaks
certain mem size limits. When you enable the rn0 device, be sure to also
change (if needed) its port and iomem parameters.
In addition to enabling the rn0 device, be sure you also do the following:
o Enable the network option GATEWAY (this enables IP forwarding as
needed by an Internet gateway machine).
o Talk to your Internet service provider and find out what type
of protocol their routers expect. Almost 100% certain that they
will say PPP. If so, enable the PPP option; in this case you do
NOT need the CISCO_HDLC option (use this option only if you need it).
Follow the BSDI instructions and build a new kernel with the rn0 device
enabled.
Boot New Kernel; Basic Testing
------------------------------
Prior to booting with the new kernel, you should connect the N1 to the
termination equipment you plan on using, and turn on that termination
equipment.
Reboot the machine; you should see rn0 in the device list. If you miss it, use
dmesg to verify that the kernel found the N1 at the desired base I/O,
interrupt, and maddr range.
Once the machine reboots, the CSU/DSU RX data light should be on. Now use
ifconfig to enable the rn0 interface. Note that if the N1 isn't connected to
line termination device (CSU/DSU), or if cable is bad/incorrect, you'll get
"rn0 timeout" messages displayed on the console.
Use the following basic ifconfig line to enable the interface:
ifconfig rn0 inet Host_IP_Address RemotePort_IP_address
At this point the TX data light on the CSU/DSU should come on. If it does,
basic functionality is OK, but don't try much else until you read the following
sections. At this point you can use two IP addrs from your assigned block,
however, the CSU/DSU should NOT be connected to the leased line.
I used a Dowty DCP3080 CSU/DSU; the only setting I had to change was to enable
the V.35 interface (instead of the serial interface). Note that in making this
type of change, you'll probably have to power cycle/reset the CSU/DSU.
Note that the rn0 driver doesn't seem to support DTR, so the CSU/DSU DTR light
doesn't come on (and neither the DSR light if CSU/DSU configed so that DSR
follows DTR). This is not a problem; things will work just fine (at least with
my equipment).
Setting Up BSDI as Router
=========================
The basics steps are as follows:
o configure the rn0 interface using ifconfig
o test connection to rn0 interface
o probably want to get gated as IP router daemon
o set default route on other hosts
Ifconfig/Basic Test
-------------------
To configure the rn0 interface using ifconfig, you'll need the following Internet
connection parameters:
+ the IP address for the rn0 interface
+ the IP address of the interface at the service provider's end
+ the netmask and broadcast values for the rn0 interface
The IP addresses, netmask, and broadcast values you get from the Internet
service provider, or if you have an existing Internet connection, you can login
to your site's router (if you have the passwords) and dump the config data.
If you have no idea what this means, get the info from your Internet provider.
The above parameter's are crucial to proper connection function. Don't try
anything until you have them; things will not work.
For example, my ifconfig (in /etc/netstart) looks as follows (NOTE link0!):
ifconfig rn0 inet 131.119.67.134 131.119.67.133 link0 \
netmask 255.255.255.252
The first IP address is the IP address of the rn0 interface (mentioned as
$hostname in template form in /etc/netstart); the second IP address is the
remote port (of a Cisco router in this case); in /etc/netstart rn0 template
this is the __remotehost__. Though it might be nice to use names instead
of IP addrs, you'll probably have trouble with names (known bsdi isssue).
If the connection is PPP, you MUST specify the "link0" interface option!
If the connection is CISCO_HDLC, you don't need link0.
Netmask is per service provider instructions; broadcast uses default
131.119.255.255 which is fine, again per service provider info (see
ifconfig man page).
Before editing /etc/netstart, enter the ifconfig command manually. Verify
TX data light goes on in the CSU/DSU. Don't be to anxious; if routed or other
router daemon is running, kill them before entering the ifconfig command
and verify that your routing table is minimal (netstat -r or -nr).
Now create a default route (route add default IP_addr); use the IP address
of the machine with the N1.
Once the default route is created, you should have connectivity. Test DNS
resolution, etc. Things should work fine. If not, use tcpdump to view
activity (tcpdump -i rn0). Note that from this point on exactly what happens
depends a lot on you Internet service provider. At a minimum I would hope that
tcpdump would show RIP requests, and maybe SNMP requests. This would indicate
that your connection is functioning and accepting packets. Outbound packets
can be verified with something as simple as a ping; this will show that IP
forwarding is working.
See next section for some Internet related details. If things worked, you
can edit your /etc/netstart file and add the ifconfig line. However, at this
point, you should comment out any 'route add default' command; see next
section.
Router Software
---------------
To turn the BSDI/N1 machine into a routc/netstart) looks as follows (NOTE link0!):
ifconfig rn0 inet 131.119.67.134 131.119.67.133 link0 \
netmask 255.255.255.252
The first IP address is the IP address of the rn0 interface (mentioned as
$hostname in template form in /etc/netstart); the second IP address is the
remote port (of a Cisco router in this case); in /etc/netstart rn0 template
this is the __remotehost__. Though it might be nice to use names instead
of IP addrs, you'll probably have trouble with names (known bsdi isssue).
If the connection is PPP, you MUST specify the "link0" interface option!
If the connection is CISCO_HDLC, you don't need link0.
Netmask is per service provider instructions; broadcast uses default
131.119.255.255 which is fine, again per service provider info (see
ifconfig man page).
Before editing /etc/netstart, enter the ifconfig command manually. Verify
TX data light goes on in the CSU/DSU. Don't be to anxious; if routed or other
router daemon is running, kill them before entering the ifconfig command
and veddress and everything worked fine from that machine. The problem was with
the other hosts using the gateway machine as the default route. Other hosts
packets went out; replies never made it back. Fun to verify this using
traceroute and tcpdump.
My service provider would not help with routed; they are familar with gated
and basically said "use it".
Here's my most favorable experience with gated:
1. ftp to gated.cornell.edu; cd pub/gated
2. get gated-R3_0_2.tar.Z (make sure set transfer mode to bin!)
3. Uncompress/untar; read README; follow instructions to build
4. Use the minimal gated.conf file that says "rip yes ;"
5. Install gated.conf in /etc; gated binary in /sbin (NOTE: as built gated is
HUGE; might not fit in /sbin on root partition. Do a 'strip gated' to
remove symbolic info and reduce to reasonable size.)
6. Edit /etc/netstart to say NO to routed
7. Edit /etc/rc to enable gated
Once I rebooted with gated; everything worked!
NOTE: Gated README file cautions that for RIP to work, kernel must support
UDP checksums. By default, the BSDI 1.0 kernel does support UDP checksums; its
all set to work.
To finish up, set the default route on all other hosts to point to the gateway
machine. Note that if you're switching from a router to a bsdi machine, you
could use the IP addr of the router's ethernet interface as the IP address
of the bsdi machine. I did not this this so I coauld bounce back and forth
between router connection and bsdi connection until things were debugged.
Summary
-------
Hopefully the basic steps in this document are useful, however they are not
a susbstitute for common sense. Be creative, especially in initial stage.
For example, as long as CSU/DSU not connected to Internet line, and ifconfig
with junk IP addrs could show that things are basically functional by causing
the TX data light to go on.
Also keep in mind a cooperative Internet service provider is needed. Good
luck in this sense.
-michael
--
Michael Galassi nerd@percy.rain.com