*BSD News Article 50950


Return to BSD News archive

Newsgroups: comp.unix.bsd.freebsd.misc
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!msunews!uwm.edu!math.ohio-state.edu!howland.reston.ans.net!tank.news.pipex.net!pipex!in2.uu.net!spcuna!ritz!bet
From: bet@ritz.mordor.com (Bennett Todd)
Subject: Re: "An HTTP software server can pummel a CPU..."
References: <gary-1309951409030001@bhb17.acadia.net> <438u8f$cok@kadath.zeitgeist.net> <MICHAELV.95Sep14230434@MindBender.HeadCandy.com> <43bqgn$861@atusks02.aut.alcatel.at>
Organization: Mordor International BBS - Jersey City, NJ
Date: Fri, 15 Sep 1995 14:36:02 GMT
Message-ID: <DEyB82.A2B@ritz.mordor.com>
Lines: 41

In article <43bqgn$861@atusks02.aut.alcatel.at>,
Marino Ladavac <ladavac@aut.alcatel.at> wrote:
>Michael L. VanLoon (michaelv@MindBender.HeadCandy.com) wrote:
>: I'll bet you could still get decent user response if you had a medium-speed
>: CPU [...] and (the key ingredient) put all of the database stuff on one
>: or more drives connected to a PCI SCSI controller (or two), then put a
>: completely separate PCI SCSI controller in the machine and attach the user
>: drives to that one. I'll bet a modern OS would be usably responsive under
>: such conditions.
>[...]
>P.S: of course that what Michael suggests would work; after all, that's the
>common practice from the dawn-a-time of mainframes [...].

Change that to _should_ work, and then I think it's a fine statement of
fact.

One of the big-name academics released a Unix-like OS only a few years
ago. He proudly boasted what a _lovely_ architecture it had; it was
modular; it was message-passing; it was all ready to go distributed; it was
perfect-in-ev-er-y-way. Also, it could only service one disk I/O request at a
time --- the disk device driver was non-reentrant and single-threaded. Better
still, this paragon of the ivory towers, author of many major CS textbooks,
books, boasted of what an architectural gem his little OS was, and even had
the brass to disparage, in public, Linus Benedict Torvalds for having produced
Linux --- he indicated that he'd have failed Linus if Linus had done that for
credit in one of his classes.

So yes, it definitely should work, and that has been the canonical approach
to this problem for better than 30 years --- but I wouldn't swear that it
_will_ work until it's tested, as you can never tell what might be festering
in the mind of some potential kernel developer:-).

I'd altogether expect FreeBSD, Linux, NetBSD, and BSDI to get this right;
I'd figure most of the other Unixes would do OK, though Solaris 2 might
require a P120 and 256MB of memory to perform adequately --- but I'd expect
Minix to fail to start a new disk I/O until the previous one is completely
finished.
-- 
-Bennett
bet@mordor.com
http://www.mordor.com/bet/