home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.sys.novell:11193 bit.listserv.novell:20011 comp.arch.storage:899 comp.periphs:1729
- Path: sparky!uunet!haven.umd.edu!darwin.sura.net!spool.mu.edu!uwm.edu!psuvax1!rutgers!concert!samba!nuntius
- From: tmorris@sph.unc.edu (Thomas P. Morris)
- Newsgroups: comp.sys.novell,bit.listserv.novell,comp.arch.storage,comp.periphs
- Subject: Re: Server Fault Tolerant H/W Sources Sought
- Message-ID: <1993Jan6.233950.23303@samba.oit.unc.edu>
- Date: 6 Jan 93 23:39:50 GMT
- References: <199301042217.AA07305@tuna.wang.com>
- Sender: usenet@samba.oit.unc.edu
- Organization: UNC School of Public Health, CAIS
- Lines: 92
- Nntp-Posting-Host: mersenius.sph.unc.edu
- X-Useragent: Nuntius v1.1
-
- In response to my query for sources for fault tolerant hardware for our
- NetWare file server.....
- In article <1993Jan05.142247.27619@rchland.ibm.com> Del Cecchi,
- cecchi@rchland.vnet.ibm.com writes:
- >
- >1.5 GB is only one drive, worth about 2500 or 3000 bucks. Do you really need
- >the fault tolerance? MTBF of modern drives is in the hundreds of thousand
- hours
- >range. RAID seems like swatting flies with a sledge hammer for this app.
- >
-
- Yes, Del, I -really- need at least -some- form of more-or-less
- automatic-failover
- fault tolerance, though you are right that RAID seems like "swatting flies."
- RAID is only one of the options I am looking at; what I am trying to do is to
- perform a costs-benefits analysis vis a vis different levels of fault
- tolerance in hardware and software. Some of
- the current options available (duplexing, mirroring, RAID, the ability to
- "swap" out
- components without shutting down the system) are really -not- that expensive,
- and
- provide for nearly 100% availability during scheduled uptime.
-
- If you read the entirety of my original article, I wrote:
-
- >[...some material omitted...]
- >This server serves ~90 DOS & Windows clients, 60-70 simultaneously attached
- >including 6-7 HP printers with JetDirect print server cards. We have NW4MAC
- support
- >loaded and will shortly attach ~30 Mac clients and 3-4 AppleTalked
- LaserWriters. >Longer term, we expect to have > 200 total client machines in
- the next 6-12 months,
- >and up to 300 or more client machines beyond the next 12 months, and up to 30
- more
- >printserver-style queues. In terms of storage needs, I would expect to need
- to
- >upgrade to ~1.5Gb storage in the next year (not including mirror or duplex
- drives),
- >with up to 3Gb needed in the longer term.
- >[...]
-
- For my "client" users, this server is the -only- place from which they can
- launch
- their word processors, spreadsheets, comunications programs, etc. It is also
- the
- =only= place to which they can print their results. My main concern,
- then, is =not=
- to piss off the >200-300 client users by having =any= form of "major"
- unscheduled downtime.
-
- If I have -no- form of faul tolerance, or if if -do not- keep a "hot spare"
- disk and/or
- server CPU available, then what do I do =when= (not -if-) my CPU or disk
- controller
- or a disk fails? The Manufacturers' rated MTBF ratings notwithstanding,
- what is
- =most= important to me (and to my clients) is the MTTR (mean time to repair).
- A
- rated Mean Time Between Failure (MTBF) is just that: within that rated time,
- one
- half (50%) of components of this type =will= fail. In some shorter period of
- time,
- some smaller percentage of these components =will= fail.
-
- If I have to wait a week to replace a CPU or a disk, first I lose my clients,
- and then I
- lose my job. If I setup a server so that single or even multiple components
- can fail,
- and only barely affect my clients, my clients are happy, and I am happy. If
- these
- components can fail, and if I can swap in replacement hardware without having
- to
- "down" my server, or if I only have to "down" the server for a very brief time,
- I'm even happier, and so are my clients. I'm happy -because- they are happy,
- and
- because I can keep my job.
-
- Now, back to the mention of "swatting flies." Let's look at some ballpark
- numbers.
- A 3-1Gb drive, 3-controller, RAID 5 implementation with ~2Gb of available user
- disk
- space from one vendor runs in the $11,000 range. A standard NetWare "disk
- duplexing"
- configuration with 4-1Gb drives and 2 controllers costs in the $9,000 range.,
- about
- 20% less expensive. If I understand RAID 5 correctly (which I probably don't
- :-),
- expanding this configuration to 3Gb of available space by adding 1 additional
- 1Gb
- drive, only costs me ~$3500 more; doing the same upgrade with duplexing costs
- me
- ~$5000 more.
-