News Group Posts

smp cpu2 slow

From: Bela Lubkin <belal@sco.com>
Subject: Re: Any Known Issue with SMP Ver 1.1.1Ga for OpenServer 5.0.6?
Date: Sat, 2 Nov 2002 06:30:36 GMT
References: <ciHw9.166205$%h2.53807@news02.bloor.is.net.cable.rogers.com> 

JP wrote:

> I have posted some messages regarding some unexplainable %sys activities in
> SCO even when system was not running any tasks.  It's been a few months but
> the problem was still not resolved.
> 
> The system has two Xeon processors running SMP 1.1.1Ga.  The problem will
> disappear if I keep rebooting and run sar to verify.  There is another way
> to bring the %idle back to 100.  The trick is to deactivate the second CPU.
> That way I can see %sys shown as zero and %idle go back to 100.
> 
> I have phsyically swapped the 2 CPU's but it did not make any difference.
> No matter which CPU take turn as the "second" CPU, every time I disable the
> second CPU, I can see a better performance from the sar.  I cannot help
> wondering if there is any bug or incompatibility between SMP and the
> hardware.  I am using HP NetServer LH3000 U3 (HP part no. P2482b).

I'm having a lot of trouble following the discussion since you keep
posting from different accounts, with different subject lines, and
including either way too much or no context at all...  Some of what I'm
going to say here might not be accurate due to lack of that context.

The "smoking gun" symptom that you're seeing -- a couple of percent of
sys time on the 2nd CPU -- is rather subtle.  Most people would never
notice it, and you wouldn't be concerned about it either if there wasn't
a secondary symptom.  You say that your users complain about performance
when the system is in the "bad" state.  You need to characterize _that_
performance problem, because it's is _not_ caused by the 2% sys time.  I
am not going to accept that your users are so performance-sensitive that
they notice and complain about a 1-2% difference.

I'm not saying that the two aren't tied together.  But you must write a
description of the actual performance issues your users see, which must
be something other than "their jobs only run at 98-99% of normal speed".

No, there is no known issue that fits the description you've been giving,
as far as anyone can tell.

You've also made inconsistent statements which are making it very hard
to follow the problem.  Previously you've shown how you reboot and
sometimes see 2% sys time, sometimes see 0%.  You said that the users do
not complain on a "good" boot where %sys is 0.  In fact you seem to say
so in this message, but then you also say "every time I disable the
second CPU, I can see a better performance from the sar".  That's
confusing the matter.  If you mean "with two CPUs I see some bad boots
and some good; with one, all boots are good" -- say that.

From the symptoms you describe, I am _guessing_ at a possible cause.  It
_sounds_ like the 2nd CPU is running at a greatly reduced speed.  The
"smoking gun" symptom would happen because the CPU is never entirely
idle.  Both CPUs take 100 timer ticks per second, for instance.  With a
normally functioning full speed CPU, handling those ticks probably takes
less than .1 % of a CPU.  Now suppose the CPU was running 20x slower,
for some reason.  It would still take as many clock cycles to handle
each timer tick, but now there are 1/20 as many total clock cycles per
realtime second, so now the CPU is 2% busy.

To users, this would show up as some sort of erratic slow execution.
Each process starts out on one CPU or the other and tends to stick to
the same CPU, but can also migrate depending on system activity.
Running the same job repeatedly would result in varying runtimes.

And of course this is only happening during some boots.

To summarize, this is my _guess_ based on the symptoms:

  - some boots are fine, both CPUs work at full speed
  - some boots are bad, CPU 2 runs at a significantly reduced speed
    (for an as-yet unknown reason)
  - when CPU 2 is running slowly, your users complain about performance,
    and you notice it in `sar` because just handling the timer ticks
    takes enough CPU to be noticable

Here's a shell script which may help diagnose this.  It starts up one
process per CPU, then in each process, times a simple spin loop several
times.  If the system is otherwise idle, each process will end up
running on a separate CPU.  It should be quite obvious in the output if
one process is running significantly faster than the other.  In that
case, _which_ process runs faster may change over the course of a run,
but you'll still see that something weird is happening.

On an idle system, what you _should_ see is that each loop takes about
the same time (with maybe up to 10% variation), and the entire set of
loops run by each process should end at about the same time.  If one CPU
is running significantly faster then you'll see some loops that take a
lot less time, and one process may finish long before the other.

VERY IMPORTANT: run this at least once on a "good" 2-CPU boot and once
on a "bad" 2-CPU boot.  The point is to compare behavior between the two
states.

>Bela<

=============================================================================

#!/bin/sh

LOOPS=10        # how many times to run the outer loop
SPINS=2000000   # adjust this manually so each loop takes about 1 second

procs=--
trap 'kill -1 $procs >/dev/null 2>&1; exit' 1 2 3 15
ncpu=`LANG=C uname -X | awk '/NumCPU/ { print $3 }'`
proc=0
while [ $proc != $ncpu ]; do
  proc=`expr $proc + 1`
  loop=1
  while [ $loop -le $LOOPS ]; do
    echo Process $proc loop $loop: `/bin/time awk 'BEGIN { for(i=0; i<'$SPINS'; i++) ; }' 2>&1`
    loop=`expr $loop + 1`
  done &
  procs="$procs $!"
done
wait
 

Comments?

Jump to Comments

Many of the products and books I review are things I purchased for my own use. Some were given to me specifically for the purpose of reviewing them. I resell or can earn commissions from the sale of some of these items. Links within these pages may be affiliate links that pay me for referring you to them. That's mostly insignificant amounts of money; whenever it is not I have made my relationship plain. I also may own stock in companies mentioned here. If you have any question, please do feel free to contact me.

Specific links that take you to pages that allow you to purchase the item I reviewed are very likely to pay me a commission. Many of the books I review were given to me by the publishers specifically for the purpose of writing a review. These gifts and referral fees do not affect my opinions; I often give bad reviews anyway.

We use Google third-party advertising companies to serve ads when you visit our website. These companies may use information (not including your name, address, email address, or telephone number) about your visits to this and other websites in order to provide advertisements about goods and services of interest to you. If you would like more information about this practice and to know your choices about not having this information used by these companies, click here.