Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!news.mira.net.au!news.mel.connect.com.au!munnari.OZ.AU!news.ecn.uoknor.edu!news.wildstar.net!news.ececs.uc.edu!newsrelay.netins.net!newsfeed.dacom.co.kr!arclight.uoregon.edu!usenet.eel.ufl.edu!news.mathworks.com!nntp.primenet.com!news1.best.com!nntp1.best.com!flash.noc.best.net!not-for-mail
From: dillon@best.com (Matthew Dillon)
Newsgroups: comp.os.linux.misc,comp.unix.bsd.freebsd.misc,comp.infosystems.www.misc
Subject: Re: Unix too slow for a Web server?
Date: 25 Sep 1996 10:18:26 -0700
Organization: Best Internet Communications, Inc. (info@best.com)
Lines: 55
Distribution: world
Message-ID: <52bph2$9qo@flash.noc.best.net>
References: <323ED0BD.222CA97F@pobox.com> <Pine.BSF.3.91.960923200853.12260C-100000@dyslexic.phoenix.net> <Dy91KF.IMA@interactive.net> <52am8g$fvs@nntp1.u.washington.edu>
NNTP-Posting-Host: flash.noc.best.net
Xref: euryale.cc.adfa.oz.au comp.os.linux.misc:131598 comp.unix.bsd.freebsd.misc:27968 comp.infosystems.www.misc:44091
:In article <52am8g$fvs@nntp1.u.washington.edu>,
:Trent Piepho <xyzzy@u.washington.edu> wrote:
:>In article <Dy91KF.IMA@interactive.net>,
:>Chris Mauritz <ritz@onyx.interactive.net> wrote:
:>>In comp.unix.bsd.freebsd.misc Geoff Mohler <gemohler@phoenix.net> wrote:
:>>: On Mon, 23 Sep 1996, Jordan K. Hubbard wrote:
:>>
:>>: > *grin*. I think you really need to talk to BEST Internet, over at
:>>: > http://www.best.net - they might even have a few Challenge machines to
:>>: > sell at reasonable prices, having already tried this! :)
:>>
:>>: I picked up a Challenge-L 128Mb/2G for under $70k, gonna add a mirror
:>>: array to it and a second Ethernet.
:>>
:>>For $70k, you could buy 10 fully configured Pentium Pro 200mhz boxes
:>>that would (in aggregate) probably be able to service connections at a
:>>rate one or two orders of magnitude greater.
:>
:>But how would you share the data between them? Unless you buy a RAID array
:>for each machine, you'll have to use NFS or something. If you factor in
:>20 gigs of drive for each ppro, you'll have a hard time buying 10.
:>
:>--
:>|Gazing up to the breeze of the heavens \ on a quest, meaning, reason |
:>|came to be, how it begun \ all alone in the family of the sun |
:>|curiosity teasing everyone \ on our home, third stone from the sun. |
:>|Trent Piepho (xyzzy@u.washington.edu) -- Metallica |
I would avoid NFS like the plague... doesn't matter *how* fast your raid
box is, you are still massively limited by both aggregate network bandwidth
and transactional network latency. 100Base-T can only do about
5 MBytes/sec, and FDDI only about 9 MBytes/sec, whereas a
fast-wide-differential or ultra-fast SCSI can do 20 MBytes/sec. You just
stick a couple of SCSI controllers in your pentium pro and you have
SCSI bandwidth that is an order of magnitude greater then anything you
could possibly get over a network... on *each* box, verses hitting up
against an aggregate transactional limit or a network
transactional/bandwidth limit with an NFS mounted raid box or host.
NFS is also not necessarily as stable when compared to a directly mounted
filesystem, especially under heavy loads. I have yet to see *anyone*
be able to scale up a machine using NFS. The problem is almost the same
problem one faces running a large news server.
Now, of course, that means shared filesystems are out. You either have
to separate the material, or do mirroring. If you have control over the
information, mirroring actually isn't that bad if you are careful in
your implementation.
-Matt
--
Matthew Dillon Engineering, BEST Internet Communications, Inc.
<dillon@best.net>
[always include a portion of the original email in any response!]