LKML Archive on
help / color / mirror / Atom feed
From: "Asbjørn Sannes" <>
To: Nick Piggin <>
	Linux Kernel Mailing List <>
Subject: Re: Unpredictable performance
Date: Mon, 28 Jan 2008 10:12:24 +0100	[thread overview]
Message-ID: <> (raw)
In-Reply-To: <>

Nick Piggin wrote:
> On Saturday 26 January 2008 02:03, Asbjørn Sannes wrote:
>> Asbjørn Sannes wrote:
>>> Nick Piggin wrote:
>>>> On Friday 25 January 2008 22:32, Asbjorn Sannes wrote:
>>>>> Hi,
>>>>> I am experiencing unpredictable results with the following test
>>>>> without other processes running (exception is udev, I believe):
>>>>> cd /usr/src/test
>>>>> tar -jxf ../linux-
>>>>> cp ../working-config linux-
>>>>> cd linux-
>>>>> make oldconfig
>>>>> time make -j3 > /dev/null # This is what I note down as a "test" result
>>>>> cd /usr/src ; umount /usr/src/test ; mkfs.ext3 /dev/cc/test
>>>>> and then reboot
>>>>> The kernel is booted with the parameter mem=81920000
>>>>> For the results vary from (real time) 33m30.551s to
>>>>> 45m32.703s (30 runs)
>>>>> For with nop i/o scheduler from 29m8.827s to 55m36.744s (24
>>>>> runs) For also varied a lot.. but, lost results :(
>>>>> For only vary from 34m32.054s to 38m1.928s (10 runs)
>>>>> Any idea of what can cause this? I have tried to make the runs as equal
>>>>> as possible, rebooting between each run.. i/o scheduler is cfq as
>>>>> default.
>>>>> sys and user time only varies a couple of seconds.. and the order of
>>>>> when it is "fast" and when it is "slow" is completly random, but it
>>>>> seems that the results are mostly concentrated around the mean.
>>>> Hmm, lots of things could cause it. With such big variations in
>>>> elapsed time, and small variations on CPU time, I guess the fs/IO
>>>> layers are the prime suspects, although it could also involve the
>>>> VM if you are doing a fair amount of page reclaim.
>>>> Can you boot with enough memory such that it never enters page
>>>> reclaim? `grep scan /proc/vmstat` to check.
>>>> Otherwise you could mount the working directory as tmpfs to
>>>> eliminate IO.
>>>> bisecting it down to a single patch would be really helpful if you
>>>> can spare the time.
>>> I'm going to run some tests without limiting the memory to 80 megabytes
>>> (so that it is 2 gigabyte) and see how much it varies then, but iff I
>>> recall correctly it did not vary much. I'll reply to this e-mail with
>>> the results.
>> 5 runs gives me:
>> real    5m58.626s
>> real    5m57.280s
>> real    5m56.584s
>> real    5m57.565s
>> real    5m56.613s
>> Should I test with tmpfs aswell?
> I wouldn't worry about it. It seems like it might be due to page reclaim
> (fs / IO can't be ruled out completely though). Hmm, I haven't been following
> reclaim so closely lately; you say it started going bad around 2.6.22? It
> may be lumpy reclaim patches?
Going to bisect it soon, but I suspect it will take some time
(considering how many runs I need to make any sense of the results).

Asbjorn Sannes

  reply	other threads:[~2008-01-28  9:12 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-01-25 11:32 Asbjorn Sannes
2008-01-25 14:00 ` Nick Piggin
2008-01-25 14:31   ` Asbjørn Sannes
2008-01-25 15:03     ` Asbjørn Sannes
2008-01-26  0:38       ` Nick Piggin
2008-01-28  9:12         ` Asbjørn Sannes [this message]
2008-01-25 17:16 ` Ray Lee
2008-01-25 20:49   ` Asbjørn Sannes
     [not found]     ` <>
2008-01-28  9:02       ` Asbjørn Sannes

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \ \ \ \ \ \
    --subject='Re: Unpredictable performance' \

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).