Stress testing your machine -- what program?

Lennart Sorensen lsorense-1wCw9BSqJbv44Nm34jS7GywD8/FfD2ys at public.gmane.org
Tue Apr 19 15:07:59 UTC 2011


On Tue, Apr 19, 2011 at 08:00:20AM -0700, William Park wrote:
> > From: Lennart Sorensen <lsorense-1wCw9BSqJbv44Nm34jS7GywD8/FfD2ys at public.gmane.org>
> 
> >
> > On Mon, Apr 18, 2011 at 07:46:46PM -0400, William Park wrote:
> > > When there  is nothing going on, hdparm, dd, kernel compile, etc., gives
> > > expected  numbers.  When things are going on, however, my machine
> > >  crawls/freezes/stutters in a way that is not linear in terms of system
> > >  load.  I looked at swap, and it's not even using swap.
> > > 
> > > I  found StressLinux from <distrowatch.com>... downloading...
> > 
> > So which  kernel version?  Some filesystems and some kernel versions have
> > pretty  bad bahaviour under some types of load.
> 
> kernel-2.6.38.2
> ext4

That one should behave fairly well.  ext3 with 2.6.29 and earlier was
really bad.

Of course if you simply have too much disk I/O there isn't much to do
about it.  Programs that do sync a lot are not good for performance
of course.  ext3 has unfortunately made a lot of programs bahave really
badly over the years which will take some time to undo.

hdparm tests are not meant to be done with other things.  It is a good
test of raw read speed, but that's it.  Not a stress test and not to be
run with other things.

kernel compile is a nice cpu/memory test, but doesn't do that much
disk I/O.  These days you could even run it from a ramdisk to elliminate
disk I/O entirely from a performance test.

dd like hdparm is rather direct and brutal, and not really a good test
of filesystem performance, only raw disk I/O.

Of course if you run software raid5 then that does require CPU to handle,
especially if you haven't aligned your data properly for it.

-- 
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists





More information about the Legacy mailing list