Thanks to everyone. I believe I just need more memory and a 100 MB
backplane in my network hub (along with FastEthernet). Here are all
of the responses.
In article cop@maze.dpo.uab.edu, det@dukebar.crml.uab.edu (Don Turrentine) writes:
> Hi all,
>
> I have a multiple CPU (4) SparcServer 1000 and it seems real slow
> at times. I sometimes have the sneaky feeling that one CPU is doing
> all the work and the other three are leaning on crutches. Any ideas
> on how to tell if the load is getting distributed?
>
> Thanks,
>
> Don
>
How about mpstat e.g.
{/home/andrewc/games/doom} $ mpstat
CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl
0 10 0 0 127 26 102 27 0 0 0 258 13 3 1 83
This is on my machine (which is a uni-processor) idl will show you your idle
CPU's.
Remember that if you are running a single CPU intensive task an MP machine
will not give you a kick unless it is multi-threaded. However you should
see performance gains/improvements in a multi-user multi-job environment
without doing anything.
It is likely that your bottleneck may not be CPU related, look at the I/O and
network stats with iostat -x 5, and netstat -i.
cheers,
Tim Addison, Senior SE - Cambridge UK
Check out mpstat (you may want to check out vmstat iostat & sar as well).
-chris
Try mpstat. There were problems in poll (and select, which uses poll)
in 2.3 and earlier. Since most network software uses select, this
would tend to limit the extent to which a system doing lots of network
stuff would use multiple processors. The worst case seems to be a
time-sharing system of the kind that universities typically run.
These problem are mostly fixed in 2.4, though some work is also
being done in 2.5.
What version of the OS are you using? If it's Solaris 2.4, there's a patch for
performance improvement on SC1000s - 102119-01. If not, have you checked for
other possible bottlenecks (i/o, network, etc.)? If you can get a hold of it,
the book "Sun Performance and Tuning" by Adrian Cockcroft (the author of the
white paper on the same topic) is a good reference.
Solaris 2.3? I understand 2.4 fixes MP problems. We have three CPU (4)
SS1000's and see the same behavior at times. Look at your system time.
Likely high. Something about polling. Upgrade to 2.4 and let me know if
it fixes it :)
-Rich Auletta
mpstat will tell you what each CPU is doing - but
you may need to log a call with your local answercentre
to understand what all the fields mean and exactly what
they are trying to tell you!
-- Bede Seymour Customer Services Group Sun UK AnswerCentreTake a look at "mpstat"
Morgan Herrington
-- Don Turrentine, Information Systems Specialist UAB - Cardiac Rhythm Management Laboratory det@dukebar.crml.uab.edu
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:17 CDT