observations at periodic intervals

observations at periodic intervals

[José Simão]

Dear user of linux-perf-users@vger.kernel.org,

When using the -I option of 'stat' to periodically collect the value
of performance counters (e.g. perf stat -e cache-misses -I 500 cmd)
the reported value at each internal (500ms in the example) is how much
the selected counter increased since the last observation?

I assume it is because the values are not monotonically increasing.

Best,
Jose

Re: observations at periodic intervals

[Arnaldo Carvalho de Melo]

Em Fri, Nov 06, 2015 at 09:06:14PM +0000, José Simão escreveu:
> Dear user of linux-perf-users@vger.kernel.org,
> 
> When using the -I option of 'stat' to periodically collect the value
> of performance counters (e.g. perf stat -e cache-misses -I 500 cmd)
> the reported value at each internal (500ms in the example) is how much
> the selected counter increased since the last observation?
> 
> I assume it is because the values are not monotonically increasing.

From 'perf stat --help' /-I

       -I msecs, --interval-print msecs
           Print count deltas every N milliseconds (minimum: 100ms)
             example: perf stat -I 1000 -e cycles -a sleep 5

This is also available via 'man perf-stat'

So, yes, your observation is correct, those are deltas.

- Arnaldo