Tomas,

Another point here, is that cpu_clk_unhalted:ref as its name indicates
only counts
when the CPU is not in halted state, i.e, not idle. Thus, it is not
correlated to wall
clock time unless you are sure the CPU never goes idle.

By default, pfmon ONLY counts user level events, make sure you pass -u -k to
measure at both user + kernel.



On Mon, Apr 21, 2008 at 10:45 AM, stephane eranian
<[EMAIL PROTECTED]> wrote:
> Tomas,
>
>
>  On Mon, Apr 21, 2008 at 10:23 AM, Tomas Tuma1 <[EMAIL PROTECTED]> wrote:
>  >
>  >  Vince,
>  >
>  >
>  >  > I'm not sure if it's causing this particular problem for you, but check
>  >  > the intel errata.  There's one for this perf counter that makes the 
> count
>  >  > vary with system frequency on some machines.  A core2 machine we have
>  >  > exhibits this erratum.
>  >
>  >  Thank you, it is important to know that.
>  >
>  >
>  >  >
>  >  > I don't think that would cause a 100x difference though.
>  >  >
>  >
>  >  The output with sampling period set to 1000 ms (1s) looks like this
>  >  (the first event is the CPU_CLK_UNHLATED:REF, the second event are cache
>  >  misses - not important):
>  >
>  >   CPU0   61904654   77838
>  >   CPU1   32798577   47473
>  >   #
>  >   CPU0  168643020  213794
>  >   CPU1   90882904  171191
>  >   #
>  >   CPU0  590366000  770770
>  >   CPU1 1172523359 1644810
>  >
>  >
>  Fluctations are not unexpected when the system is doing nothing (or at
>  least you think it is).
>  What is more important is to verify that when you have a constant
>  workload the count for
>  REF remains fairly constant and close to peak for a nonblocking
>  workload, i.e., a stupid noploop
>  program such as (pfmon/tests/noploop).
>

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
perfmon2-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel

Reply via email to