Hi,
I suspect you are running on an Intel machine in which case I believe you
could colect everything in one run. This is if the NMI watchdog is turned
off.
If you can turn it off then you have two generic counters for the cache
events and a fixed counter for instructions_retired.
That would help a bit with fluctuations. But wih cache misses, nothing is
deterministic. It depends on a lot of factors outside of your control.
On Sep 7, 2010 3:53 AM, "lulu he" <lolos...@gmail.com> wrote:
> Hi,
>
> I currently use pfmon to collect cache_miss_rate every 1M insns of
> parallel programs.
> Because my server only has 2 PMUs, so I have to run benchmark
> twice(once measure cache_references every 1M insns, twice cahce_misses
> every 1M insns) to get cache_miss_rate every 1M insns.
>
> I use the same input , and my enviroment is relatively stable.
> The problem is most cache_miss_rate I got is >1.
> I think it's because of pfmon's sampling property.
>
> Can u tell me what's the problem, and how can I avoid/reduce such
variability?
>
> Thanks.
------------------------------------------------------------------------------
This SF.net Dev2Dev email is sponsored by:
Show off your parallel programming skills.
Enter the Intel(R) Threading Challenge 2010.
http://p.sf.net/sfu/intel-thread-sfd
_______________________________________________
perfmon2-devel mailing list
perfmon2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
