NetNews Usenet Archive 1993 #1/NN_1993_1.iso/spool/comp/unix/sys5/r4/1230

Internet Message Format | 1993-01-11 | 4.0 KB

Xref: sparky comp.unix.sys5.r4:1230 comp.unix.questions:15430 biz.comp.telebit.netblazer:30 comp.protocols.tcp-ip:5922
Newsgroups: comp.unix.sys5.r4,comp.unix.questions,biz.comp.telebit.netblazer,comp.protocols.tcp-ip
Path: sparky!uunet!nntp.telebit.com!mlewis
From: [e-mail] (Mark S. Lewis)
Subject: Re: Netblazer
In-Reply-To: [e-mail]'s message of 8 Jan 93 16:37:18 GMT
Message-ID: <[e-mail]>
Sender: [e-mail]
Nntp-Posting-Host: yoyo
Reply-To: [e-mail]
Organization: Telebit Corporation; Sunnyvale, CA, USA
References: <[e-mail]> <[e-mail]>
Date: Mon, 11 Jan 1993 23:56:19 GMT
Lines: 74


> [e-mail] (Mark S. Lewis) writes:
>>> I currently attach via SLIP over a leased line using v.32bis/v.42bis
>>> modems from a Dell machine (on my end) into a smartport on a Telebit Netblazer.
>>
>>> Would I see any significant increase in throughput using this same
>>> physical connection (v.32bis/v.42bis) but replacing the Dell SLIP
>>> with a Netblazer on my end -- and connecting my machines via ethernet
>>> to the Netblazer?
>>> [...]
>>> Larry Snyder                                    internet: [e-mail]
>>
>>No I don't think you would get a significant performance increase by
>>putting a NetBlazer at both ends.  You are using relatively slow
>>serial lines limited by V.32bis modulation with whatever compression
>>you get with V.42bis.
>>
>>Mark S. Lewis  inet: [e-mail]

> I disagree.  I think you will get some increase in preformance, depending
> on how loaded your box that is doing the routing is.  Routing IP packets
> takes a certain amount of CPU over head as well as servicing the serial
> line interupts that happen every time you get a packet in.  Using 16500
> helps.

I suppose it is possible to improve performance by shifting the burden
of routing to the NetBlazer, 386BSD or whatever.  Depends on the
routing effeciency of the original system.

Also, the NetBlazer typically drives modems at 57600, while many Unix
based SLIP connections can reliably support only 19200.  In this case,
a NetBlazer or dedicated SLIP solution may help.

> A side note.  Netblazer software was based on the KA9Q software.  Enough
> work has gone into it by now that I imagine that it doesn't look like KA9Q
> any more, but I wonder if it still has the same problems that KA9Q has,
> like memory leakage and high latency. Perhaps Mark could comment on the
> Netblazer software.

As for memory leakage, we have come quite a ways.  We have made
numerous improvements to the memory allocation to help us find leaks.
At this point, there are only a few rare leaks.  In general, the
NetBlazer will stay up for months.

As for latency, I am not sure to which latency you refer.  What type
of connections are you assuming, async dial-up?  Are you talking about
round trip ping time?  Note that modems introduce significant latency.

> At the last ONEBBS con in Denver, there was a demonstration of a device
> called "Slip-on-a-Stick".  It was a box that was about 10"x6"x2" tall, and
> would do slip.  I don't know if it will do the right thing with subnetting
> as I only saw it work with one device on it.  The price was less than a
> grand.

I like that - "Slip-on-a-Stick."  Sound like the next taste sensation
:-)  Note that I know of several vendors that are developing such
dedicated networking serial devices.  It will be interesting to see
what comes down the pike.

>                             Tim

> --
>     Internet: [e-mail]     FidoNet: Tim Pozar @ 1:125/555
> Snail: Tim Pozar / KKSF / 77 Maiden Lane / San Francisco CA 94108 / USA
>                         Voice: [phone]

KKSF is my favorite jazz station.  You guys play the tastiest jazz in
the bay area.  Thanks for the good programming.

... Mark

=========-----------          Mark S. Lewis      -----------==========
inet: [e-mail]      Telebit Corp.     Telebit [phone]
uucp: uunet!telebit!mlewis                          Fax [phone]