linux-kernel - Re: [PATCH] perf trace: Simplify '--summary' output

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20131112215024.GA29761@gmail.com>
Date:	Tue, 12 Nov 2013 22:50:24 +0100
From:	Ingo Molnar <mingo@...nel.org>
To:	Pekka Enberg <penberg@....fi>
Cc:	David Ahern <dsahern@...il.com>, Pekka Enberg <penberg@...nel.org>,
	linux-kernel@...r.kernel.org, Ingo Molnar <mingo@...hat.com>,
	Arnaldo Carvalho de Melo <acme@...stprotocols.net>
Subject: Re: [PATCH] perf trace: Simplify '--summary' output


* Pekka Enberg <penberg@....fi> wrote:

> > %. That was dropped in the recent output change.
> 
> Sorry about that. Why do we show them in percentages, btw? Standard 
> deviation is usually represented in the same unit as the data to make it 
> readable.

So, the problem with output in the same unit is that it's hard to tell
'at a glance' what the real, relative stddev is:

    syscall            calls      min      avg      max stddev
    --------------- -------- -------- -------- -------- ------
    sendmsg                2    0.002    0.005    0.008 0.001
    recvmsg                2    0.002    0.003    0.005 0.0003
    epoll_wait             1    0.000    0.000    0.000 0.000

it's also harder to compare stddevs side by side if they are absolute.

The way we try to solve this in perf stat is to output:

  comet:~/tip> perf stat -a --repeat 10 --null usleep 1000

   Performance counter stats for 'system wide' (10 runs):

       0.001772475 seconds time elapsed                                          ( +-  0.58% )

Oh, btw., would anyone would like to implement a cool new 'perf time' 
subcommand, which offers time(1) look-alike output but with better 
precision and sttdev output?

  comet:~> time usleep 1000

  real    0m0.002s
  user    0m0.000s
  sys     0m0.000s

  # Mockup:

  comet:~> perf time --repeat 10 usleep 1000

  real    0m0.002s      [ +- 1.38% ]
  user    0m0.000s      [ +- 0.53% ]
  sys     0m0.000s      [ +- 0.43% ]

?

I'd use it instead of 'time' all the time ;-)

Thanks,

	Ingo
--
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/