*BSD News Article 36243


Return to BSD News archive

Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!yeshua.marcam.com!MathWorks.Com!panix!not-for-mail
From: tls@panix.com (Thor Lancelot Simon)
Newsgroups: comp.os.386bsd.questions
Subject: Re: Dual/Quad Pentium servers
Date: 28 Sep 1994 03:26:15 -0400
Organization: PANIX Public Access Internet and Unix, NYC
Lines: 41
Message-ID: <36b5qn$mv4@panix2.panix.com>
References: <Robert-CCN.jrfq@sacbbx.com> <35igq3$2no@engnews2.Eng.Sun.COM> <35kbrn$s2k@qualcomm.com> <35tsdq$bbb@engnews2.eng.sun.com>
NNTP-Posting-Host: panix2.panix.com

In article <35tsdq$bbb@engnews2.eng.sun.com>,
Jack F. Vogel <jackv@orpheus.Eng.Sun.COM> wrote:
>In article <35kbrn$s2k@qualcomm.com> bri@qualcomm.com (Brian Ellis) writes:
>|Jack F. Vogel <jackv@orpheus.Eng.Sun.COM> wrote:
>|
>|>The answer is unfortunately NO, NONE of the free BSD variants will
>|>run MP. Of course I don't know the hardware, but I owuld guess that
>|>it will RUN on the machine, it just won't initialize or do anything
>|>with the other CPUs.
>|>
>|>As someone who has worked on a couple of MP kernels, if someone
>|>wants to donate an MP motherboard (I will even foot the bill for
>|>the CPUs :-), I would LOVE to work on MP extensions to FreeBSD :-).
> 
>|Jack, if you mean you've worked on the Solaris kernel, then I'm sure
>|you know how drastically the kernel had to be re-engineered to go
>|from SunOS 4.x. Frankly, I don't think the *BSD kernels could ever
>|be made to run reliably on a symetric multiprocessing platform.

Oh, boy, does Solbourne ever have a bone to pick with you... :-)

>So the type of MP kernel I've worked on may have been more coarsely 
>grained, but the advantage is this approach is its MUCH easier to 
>implement based on a kernel that starts UP (like *BSD). That's
>why I think it would be FUN to get a coarsely-grained X86 system
>running. In fact with the primitives from Mach in FreeBSD and/or
>NetBSD it shouldn't be hard to get something finer grained
>running without much difficulty.

Besides, OS/MP is a lot less coarsely grained than SunOS 4.1.3, and it's SunOS
4-based.  So we have empirical evidence that it's not impossible to reliably
make a BSD (well, mostly) kernel do finely-grained SMP.

Even DEC did SMP with a 4.2BSD (gack!) kernel.  Of course, they had some nasty
performance problems, but a lot of that was hardware, and one has to keep in
mind also that, well, they're DEC.  :-)

-- 
Thor Lancelot Simon	                                           tls@panix.COM
-   What is hardest to accept about the passage of time is that the people who -
- once mattered the most to us are wrapped up in parentheses.    --John Irving -