[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <87k3agsla6.fsf@sejong.aot.lge.com>
Date: Wed, 23 Apr 2014 13:49:05 +0900
From: Namhyung Kim <namhyung@...nel.org>
To: Ingo Molnar <mingo@...nel.org>
Cc: Arnaldo Carvalho de Melo <acme@...nel.org>,
Jiri Olsa <jolsa@...hat.com>,
Peter Zijlstra <a.p.zijlstra@...llo.nl>,
Paul Mackerras <paulus@...ba.org>,
Namhyung Kim <namhyung.kim@....com>,
LKML <linux-kernel@...r.kernel.org>,
David Ahern <dsahern@...il.com>,
Andi Kleen <andi@...stfloor.org>
Subject: Re: [PATCHSET 0/9] perf tools: Fixup for the --percentage change
Hi Ingo,
On Tue, 22 Apr 2014 11:55:57 +0200, Ingo Molnar wrote:
> I gave it some quick testing and after fixing a trivial merge conflict
> in tools/lib/lockdep/Makefile all seems to be working fine.
Thanks for testing!
>
> But while looking at it I remembered one of my old UI complains about
> perf top and report, the hard to read nature of:
>
> Event count (approx.): 226958779
>
> the values displayed are typically way too large to be easily human
> readable. More importantly, they are also nonsensical! That we have a
> sampling interval and can sum up all the intervals sampled has very
> little meaning to the overwhelming majority of humans looking at the
> data.
>
> And printing that just spams the visual field and confuses people.
>
> People care about the quality and speed of sampling itself, not
> directly the interval of sampling (which will often be variable with
> auto-freq sampling).
You meant 'period' by 'interval', right?
There's --show-total-period option (should be equivalent to -F period
later) in perf report, so there might be people want to see the numbers
IMHO.
>
> So instead of:
>
> Samples: 42K of event 'cycles', Event count (approx.): 226958779
>
> How about only printing this in 'perf top' and 'perf report':
>
> Captured 42.1K 'cycles' event samples
>
> Note the extra decimal (which helps monitor smaller changes as well),
> and note the different wording.
>
> Thoughts?
Well, I'm okay to add the extra decimal, but it seems that it only makes
sense when the unit is 'K'..
And I think it might be worth adding filtered sample count as well if
filtering is enabled something like:
Captured 13.2K/42.1K 'cycles' event samples
Thanks,
Namhyung
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists