[Gllug] Re: Memory usage

[Peter Grandi]

>>> On Thu, 13 Oct 2005 22:06:18 +0100, Tethys
>>> <sta296 at astradyne.co.uk> said:

[ ... ]

>> On a 3GHz CPU, a user CPU time of 9 seconds is extremely
>> worrying, and the 3 seconds of system CPU time are simply
>> insane...

[ ... ]

sta296> Yes, it's a hyperthreaded CPU. But then I'm running
sta296> 2.6.11, so it should be well aware of the hyperthreading
sta296> for scheduling purposes.

Well, 2.6.11 should be good for that. But I still distrust
hyperthreading, because if the same CPU is used for other
things, bad news can happen. Also, if some other application is
running that has a pathological behaviour with hyperthreading
(e.g. a cache pollyting one), OOo can be affected too.

But it could be a few other things, but it is really hard to
imagine what could be causing that amount of CPU time. What
seems to be happening is that OOo gets scheduled across both
VCPUs, as the total CPU time is greater than elapsed, and that
can happen only if both CPUs are running in parallel. But that
is odd in itself: I have just started 'soffice -writer' (always
1.1.5 here) and it looks decidedly like a single thread.

Another conceivable possibility for the high CPU times is that
the CPU is overheating and thermal throttling is kicking in and
reducing greatly the clock speed. But that seems even less likely.

sta296> [ ... ] Again, if something was wrong with my system,
sta296> I'd expect to see the same sort of problems with all
sta296> applications, not just OO.o.

As a wild guess it had not been noticed previously that the CPU
time needed to start OOo is greater than the elapsed time, and
that the system CPU time is 10-20 times greater than it should
be, and perhaps, as another wild guess, if these extraordinary
anomalies had not been noticed perhaps others relating to the
execution of other programs may have been unnoticed too.

But of course if you know better and/or feel lucky... :-)

sta296> Note that your other suggestion (tweaking the
sta296> filesystem) will soon be put to the test. [ ... ]

While that is generally good, as I was writing it would help
towards reducing the amount of wait time (due to seeking etc.),
and as I was writing the problem with OOo that you describe
seems to be primarily very abnormally high CPU time, not high
wait times (the wait time is apparently slightly negative...)
like with everybody else. I am sorry that evidently I did not
write that clearly as to either point :-).

-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug