*BSD News Article 87444


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.Hawaii.Edu!ames!enews.sgi.com!news.sgi.com!news.sprintlink.net!news-peer.sprintlink.net!newsfeed.internetmci.com!Quza.UK.peer!uunet!in3.uu.net!192.35.48.11!hearst.acc.Virginia.EDU!aruba.odu.edu!news.cs.odu.edu!jason
From: jason@daffodil.cs.odu.edu (Jason C Austin)
Newsgroups: comp.os.linux.networking,comp.unix.bsd.bsdi.misc,comp.unix.bsd.misc,comp.sys.sun.misc
Subject: Re: Sparc vs. x86 speed (was Re: Linux vs BSD)
Date: 23 Jan 1997 20:16:39 GMT
Organization: Old Dominion University, Norfolk, VA, USA
Lines: 57
Message-ID: <JASON.97Jan23151639@daffodil.cs.odu.edu>
References: <32DFFEAB.7704@usa.net> <5c155c$p6u@raven.eva.net>
	<5c19pg$rf6@lynx.dac.neu.edu>
	<5c58n9$hcb@innocence.interface-business.de>
	<32E66DE1.7E36AB48@samart.co.th> <5c8b0o$313$1@capsicum.wsrcc.com>
Reply-To: austin@visi.net
NNTP-Posting-Host: daffodil.cs.odu.edu
In-reply-to: wolfgang@dailyplanet.wsrcc.com's message of 23 Jan 1997 10:35:04
	-0800
Xref: euryale.cc.adfa.oz.au comp.os.linux.networking:66092 comp.unix.bsd.bsdi.misc:5727 comp.unix.bsd.misc:2020 comp.sys.sun.misc:28086

In article <5c8b0o$313$1@capsicum.wsrcc.com> wolfgang@dailyplanet.wsrcc.com (Wolfgang Rupprecht) writes:
=> Ruediger Koch <rkoch@samart.co.th> writes:
=> >regularly. For the mailserver, we replaced a Sparcstation 20 with 64MB
=> >by a P133 with 32MB. The load went down from 1-2 to 0.05-0.3! Two of the
=> >fucking expensive Suns are catching dust now.

	But what was the end user performance?  Did you compare your
mail volume per hour before and after the change?  A low load average
usually means you're not getting things to the CPU fast enough to keep
it busy.  I've found "PC" type systems tend to have too much CPU for
the rest of the machine, since it looks good on an advertisement.  All
you end up with is a hurry up and wait situation which gives you a low
load average.

	A load average of 1-2 on a 1 CPU machine is just about where
you want it to be.  If it's not, you probably have an I/O bottleneck
of some sort.

=> I'd like to echo those sentiments.  For over 10 years I've been
=> running nothing but BSD machines.  For the longest of times the only
=> serious choice was Sun/SunOS (Vax/BSD always tied you to their vastly
=> overpriced peripherals).
=> 
=> I was quite suprised to hear a coworker someone claim that a
=> multi-hour compile of a large source tree (on an ss5) could be done in
=> 1/4th the time on a PentiumPro under NetBSD.  (Yea right!)  Well it
=> was true.  As a matter of fact a PPro-150 could do the compile in
=> approx 3/4 the user time that an SS20 could do it in.  That was quite
=> a shocker.  Checking the benchmarks of the two machines
=> (eg. SpecFoo's) showed a roughly similar speed ratio.

	Compile time is not a good comparision; there's just too many
variables involved.  A more optimized compiler will need much more CPU
but will produce smaller and faster assembly code.

=> I sure hope that the Sun sales/markening droids have some spin they
=> can put on these facts.  They'll need it as more and more folks
=> realise how far Sun has slipped behind commodity hardware.
=> 
=>     specint-95:
=> 	SS5:  85Mhz  1.37 
=> 	SS20: 150Mhz 3.77
=> 	PPRO: 150Mhz 4.27
=> 
=>     specmarks speed ratio:
=> 	ss5-85/ppro-150:       1:4
=> 	ss20-150/ppro-150:     1:1.13 

	As I said before, CPU is not what makes a fast machine for
most applications.  If your application is almost all CPU with low
I/O, you probably will be better off with the Pentium machine.  If
you're doing database type applications (much more common), the Sun
will leave it in the dust.  You need to get the right machine for the
job.
--
Jason C. Austin
austin@visi.net