From: Kjetil Barvik <barvik@broadpark.no>
Date: Sat, 14 Feb 2009 17:50:34 GMT
Subject: Re: [PATCH/RFC v3 7/9] write_entry(): use fstat() instead of lstat() when file is open
Message-ID: <86ab8olxg5.fsf@broadpark.no>
In-Reply-To: <498A9FD3.2020601@viscovery.net>

Johannes Sixt <j.sixt@viscovery.net> writes:

> Kjetil Barvik schrieb:
>>   And, yes, since each lstat() call cost approximately 44 microseconds
>>   compared to 12-16 for each lstat() on my Linux box, there was a little
>                                ^^^^^^^ fstat()?
>>   performance gain from this patch.
>
> This does look like a good gain. But do you have hard numbers that back
> the claim?

  OK, I have done some testing/profiling with oprofile(1), and one thing
  I found out was that my Linux kernel was built with SLUB debug, and of
  course it cost some system time to run the VM debug code.  After I
  turned this off, the total system time when down from aprox 6 to 3
  seconds for the 'git checkout -q my-v2.6.25/7' test.

  Also, from strace output each lstat() call now take around 16
  microseconds, and each fstat() call around 12 microseconds, so for
  aprox 14000 changed calls (lstat() => fstat()) the performance gain
  should now only be (* 14000 (- 16 12)) = 56 ms, compared to 467 ms,
  which I reported before.

  -- kjetil

  1) http://oprofile.sourceforge.net/about/