[rescue] OS X and dual CPU machines?

[Curtis H. Wilbar Jr.]

>Date: Sun, 8 Jun 2003 21:51:42 -0700
>Subject: Re: [rescue] OS X and dual CPU machines?
>From: "r.stricklin" <bear at typewritten.org>
>To: The Rescue List <rescue at sunhelp.org>
>Content-Transfer-Encoding: 7bit
>
>On Sunday, June 8, 2003, at 09:30  PM, N.Miller wrote:
>
>> I can't honestly tell you if the BSD sub-system is set up to take 
>> advantage of the multi-cpu facility or not, but I would expect so 
>> (seems kind of dumb if it isn't, but hell, that isn't beyond Apple).
>
>My admittedly simplistic test suggest that the kernel isn't as down 
>with SMP as it could be.
>
>The fhlushstone (the time to dd a gigabyte from /dev/zero to /dev/null 
>in 1024-byte blocks) benchmark illustrates it pretty clearly. Times are 
>for my 1.25 GHz dual G4.
>
>Single thread: 4.23 seconds.
>Two threads:  18.915 seconds.
>
>Clearly something suboptimal is happening. Contrast that to Tru64 v5.0A 
>which shows some level of threading even within the kernel... a dual 
>processor machine will run the single thread benchmark faster than a 
>single processor machine, which isn't a trait I've noticed on too many 
>other systems.

what could be happening is that there is one lock around one (or both
of those devices).... how a MP machine behaves with this test is
irrelevant to real world usage.

Back on the old SunOS 4.X days, the kernel had one lock... only one
processor could be in the kernel at once.  Going by memory I think 
Solaris 2.X had between 16 to 32 locks to start with (vague 
recollection), and has only gone up from there.

Trying to do two of the same things at the same time is not a good test
of MP...  you are likely to hit lock contention around the common
things you are trying to do... many things in the kernel can not be
truly paralleled (hence all the locks).

>
>There's a big table of results out of my lab up on the web, in case 
>you're curious how those results stack up to other systems. If you're 
>shrewd you can learn some interesting things. Not bad for a purposely 
>frivolous benchmark.

Quite frivolous... and not particularly usefull... but a bit interesting.

>
>http://www.typewritten.org/Articles/fhlushstones.html
>
>> Ugly, huh?  Netscape (7.02) has 118MB footprint with one window open.  
>> Terminal takes up 113MB or so with the default scroll back of 10K 
>> lines. :-p
>
>Except it's not using that much of your memory. It's mapped in all of 
>the backing store from the window manager (and Quartz libraries, etc., 
>etc.) which is NOT something that every process has its own copy of. 
>Basically any Quartz application running shares about 100-128 MB of 
>that VSIZE with every other Quartz application running.
>
>So in your example above, you're actually using something on the order 
>of 120 MB of real memory, and not 231 MB.
>
>A more realistic indication of memory utilization in a Quartz 
>application would be to look at the RSS (resident set size) and then 
>scale according to swap utilization.
>
>ok
>bear
>_______________________________________________
>rescue list - http://www.sunhelp.org/mailman/listinfo/rescue

-- Curt

Curtis Wilbar
Hawk Mountain Networks
rescue at hawkmountain.net

My e-mail is protected against viruses and spam by MailGuardian
                  http://www.mailguardian.net
          Top notch protection at unbelievable prices