Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.mel.connect.com.au!news.mira.net.au!news.vbc.net!alpha.sky.net!news.sprintlink.net!news-chi-13.sprintlink.net!news.fibr.net!nntp04.primenet.com!nntp.primenet.com!howland.erols.net!newsfeed.internetmci.com!newsfeed.direct.ca!nntp.teleport.com!news.serv.net!news.serv.net!michaelv
From: michaelv@MindBender.serv.net (Michael L. VanLoon)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Re: dual cpu stuff...
Date: 02 Sep 1996 09:09:21 GMT
Organization: HeadCandy Associates... Sweets for the lobes.
Lines: 99
Message-ID: <MICHAELV.96Sep2020921@MindBender.serv.net>
References: <4vcsn4$7ql@cantina.clinet.fi> <4veq47$cc@anorak.coverform.lan>
<MICHAELV.96Aug26001022@mindbender.serv.net>
<506vvp$frv@cantina.clinet.fi>
NNTP-Posting-Host: mindbender.serv.net
In-reply-to: mickey@cantina.clinet.fi's message of 30 Aug 1996 18:05:29 +0300
In article <506vvp$frv@cantina.clinet.fi> mickey@cantina.clinet.fi (Mika Ruohotie) writes:
> : multiple cpus? i heard that the new nt4 will only hand max 40%
>Well then you heard wrong. There is not one magical number that it
yes, i know it's not a single number one can hand out... but what i
heard was that nt4 is not performing too well with multiple cpus, that
it could do better too... like that something else runs the same things
better, and stuff...
Well, that is a marketing fallacy perpetuated by some companies that
would rather see Microsoft fail at industrial-strength server OS'.
Whether you like Microsoft or not, NT4 is probably the best, or at
least among the best, multi-processing OS' for Intel platforms. SCO
and UnixWare or "OK", but... they have a lot of weird quirks. Solaris
for Intel is an exercise in patience. Some day, I hope to be able to
put FreeBSD at the top of this list and really mean it.
Now, when you get to RISC platforms, you start competing with things
like Digital Unix, Solaris (ugh), Irix, etc., on really heavy-duty
Iron, not to mention special-purpose OS' like Cray's. Naturally, NT
isn't going to fare as well against those, but NT 4.0 has made a lot
of progress in this area. It's certainly _miles_ ahead of OS/2 in
this department.
>Actually, FreeBSD's implementation is very inefficient. It's very
>early alpha-quality code. This isn't to put down the people working
this is something i realise too... but still i am going to find time
to download the code and see it myself... i assume, someone please correct
me, that 'make world' is again a good way to measure this?
Hmmm... Now that I think about it, I don't think make world would be
particularly useful as a test. There isn't a lot that is really
parallel in nature with standard BSD make. Of course, you could use
the "-pipe" option for GCC, and that would parallelize some of the
work, but it wouldn't be anywhere close to keeping both CPUs busy 100%
of the time.
>sharp people in that mix. It's just that it takes time to get all the
>right pieces in place for a truly well-tuned SMP kernel, and FreeBSD
>is just getting started.
and when they're done... *smile*
When they're done, I think they will have some really great stuff.
But it's important to remember that they're just getting started,
especially if you're setting expectations.
>Well, of course it is... What does that have to do with dual Pentium
>motherboards?
my original question was do i get more performance with 1024 cache when
i use two cpus, and i think i understand here it doesnt matter, meaning
the difference is same than if i'd have one cpu and put the same amount
of cache... right?
No -- more cache should always help. It's just at what point does the
return on investment out-weigh the investment.
There's a really big jump in performance with a Unix-like box going
from 256K to 512K cache. There's a much smaller jump going from 512K
to 1024K, unless you are really pushing the machine. But there should
still be a slight improvement, since cache runs faster than DRAM, and
doesn't have to block on refresh cycles (plus can still feed the
processor when a bus-master device is using the bus).
So, in the case of a dual-Pentium system, I would say 1024KB cache
would probably be "just right". I'm curious who makes a board that
uses 1MB of cache, though? I thought the Triton stuff only went to
512K. Maybe one of the older dual chipsets?
And, also yes, dual Pentium Pros would kick-butt here. For several
reasons (not the least of which is the fact that they're just faster).
The dual Pentium Pro has the L2 cache on the "processor-side" of the
shared bus, where the dual Pentium has the L2 cache on the "bus-side"
of the shared bus. In other words, each P6 has its own L2 cache,
where the Pentiums share. At least that's the way I understand it.
It certainly would be possible to make a dual Pentium motherboard
where each chip had its own cache, but it would be a lot more
expensive. This coupled with the fact that the Pentium L2 cache runs
at bus speed (60 or 66MHz), where the Pentium Pro L2 cache runs at
processor speed (i. e. 200MHz P6 has an L2 cache that also runs at
200MHz).
I'm not going to describe all the differences. You can find out by
yourself. There's all the information you need on
http://www.Intel.com/.
--
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Michael L. VanLoon michaelv@MindBender.serv.net
--< Free your mind and your machine -- NetBSD free un*x >--
NetBSD working ports: 386+PC, Mac 68k, Amiga, Atari 68k, HP300, Sun3,
Sun4/4c/4m, DEC MIPS, DEC Alpha, PC532, VAX, MVME68k, arm32...
NetBSD ports in progress: PICA, others...
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -