2005-09-07 01:38:13

by John Richard Moser

[permalink] [raw]
Subject: Kernel profiles anyone?

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Are there any recent kernel profiles? I think from an acedemic
perspective it'd be nice to see some graphs and numbers nobody
understands showing where the longest running code paths in the kernel
occur. It might also be nice for those latency whores (*runs to the
back and raises hand*) and those who want to increase overall
performance and efficiency; then there'd be a map showing . . .
something that only kernel hackers could possibly understand or care about.

- --
All content of all messages exchanged herein are left in the
Public Domain, unless otherwise explicitly stated.

Creative brains are a valuable, limited resource. They shouldn't be
wasted on re-inventing the wheel when there are so many fascinating
new problems waiting out there.
-- Eric Steven Raymond
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDHkPIhDd4aOud5P8RAiQ4AJwPdaJAZchWNNtoO9zjz6AePpty/ACbBa/A
vX+9F3/Yuw68QtteomUXtqQ=
=faXl
-----END PGP SIGNATURE-----


2005-09-07 04:27:01

by Lee Revell

[permalink] [raw]
Subject: Re: Kernel profiles anyone?

On Tue, 2005-09-06 at 21:35 -0400, John Richard Moser wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Are there any recent kernel profiles? I think from an acedemic
> perspective it'd be nice to see some graphs and numbers nobody
> understands showing where the longest running code paths in the kernel
> occur. It might also be nice for those latency whores (*runs to the
> back and raises hand*) and those who want to increase overall
> performance and efficiency; then there'd be a map showing . . .
> something that only kernel hackers could possibly understand or care about.

There's a latency histogram option in the -rt patch set. You can pipe
the output through gnuplot and get some cool graphs. Then you can
use /proc/latency_trace to try and guess what code paths the peaks on
the graphs correspond to. The timer interrupt will be the biggest peak,
around 23 usecs last time I checked.

I haven't done it lately so it would be interesting to see the current
graphs. Someone could make a really good presentation out of it at some
kernel development conference.

Lee

2005-09-07 11:32:50

by Neil Horman

[permalink] [raw]
Subject: Re: Kernel profiles anyone?

On Wed, Sep 07, 2005 at 12:26:55AM -0400, Lee Revell wrote:
> On Tue, 2005-09-06 at 21:35 -0400, John Richard Moser wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA1
> >
> > Are there any recent kernel profiles? I think from an acedemic
> > perspective it'd be nice to see some graphs and numbers nobody
> > understands showing where the longest running code paths in the kernel
> > occur. It might also be nice for those latency whores (*runs to the
> > back and raises hand*) and those who want to increase overall
> > performance and efficiency; then there'd be a map showing . . .
> > something that only kernel hackers could possibly understand or care about.
>
oprofile is designed to do exactly this. just set it up and run the workload
you are interested in.

Regards
Neil


--
/***************************************************
*Neil Horman
*Software Engineer
*gpg keyid: 1024D / 0x92A74FA1 - http://pgp.mit.edu
***************************************************/