Mailing List Archive


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [tlug] CPU speed: reported vs. actual



Hi,

I'm not sure if the following is relevant, but ...


On Thu, Jul 14, 2011 at 10:43 PM, Attila Kinali <attila@example.com> wrote:
> On Thu, 14 Jul 2011 11:29:42 +0900
> Darren Cook <darren@example.com> wrote:
>
>> Any thoughts as to how I confirm this, and get it running at full-speed?
>> (Ubuntu 10.04, 64 bit, linux 2.6.32)
> * Is your system really cpu bound? Is it not I/O bound or memory I/O bound?
>  If you can, have a look at the cache hit statistics (don't know what tools
>  to use for this)


/usr/bin/time's man pages might be helpful.  On Debian, maybe the
number of page faults (%R) or number of times the process is swapped
out of main memory (%W) might help.  I guess if you're evaluating the
CPU, you would prefer to have very little swapping.

While you're at it, you can also compare the user, system, and real
times -- ideally, the system time should be small so that all of the
elapsed time is used by the process.

And I believe if you have MPI installed and run your program under
mpirun, there is a way to configure it so that your program runs on a
specific CPU (they might refer to it as a "slot").  I'm not sure about
this; it's been a while since I played with this part of MPI.

Ray


Home | Main Index | Thread Index

Home Page Mailing List Linux and Japan TLUG Members Links