Starting hal daemon incredibly decrease my hard drive performance

[David Zeuthen]

On Thu, 2004-11-04 at 13:57 -0500, ctwise at bellsouth.net wrote:
> On Thu, 2004-11-04 at 19:48 +0100, Sjoerd Simons wrote:
> On Thu, Nov 04, 2004 at 01:31:28PM -0500, David Zeuthen wrote:
> > > On Thu, 2004-11-04 at 17:46 +0100, Sjoerd Simons wrote:
> > > > > It's probably caused by the fact that your cdrom and your harddisk are
> > > > > sharing one ide bus. The polling of hdb may affect the hdparm
> > > > > measurement. You may try this with the udi for your cdrom:
> > > > 
> > > > Just talked to someone on IRC with basically the same problem. Turning off
> > > > polling on the cdrom drive did indeed fix the problem. 
> > > > 
> > > > Why do media checks degrade performance so much ?
> > > > 
> > > 
> > > Wait a second - the only thing we know is that polling for media change
> > > makes the numbers from hdparm lower - I'm not sure at all that
> > > performance is degraded; someone should check with other benchmarks
> > > and/or real-world tests.
> > 
> > I asked him that (as i don't really trust hdparm) and according to him
> > realworld performance also degraded.
> > 
> > He's in the CC, so i guess he should be able to give some numbers :)
> > 
> >   Sjoerd
> 
> The whole system is much more responsible after disabling the media
> check - is there a particular benchmark you would like me to run?
> 

FWIW, I've had reports today from Fedora users about this; I think it's
a IDE subsystem issue in the kernel - basically hal does this (every two
seconds)

 open("/dev/hdc", O_RDONLY|O_NONBLOCK|O_EXCL|O_LARGEFILE) = 0
 ioctl(0, CDROM_DRIVE_STATUS, 0x7fffffff) = 1
 close(0) 

Some people say that tiobench, http://sourceforge.net/projects/tiobench/
is good for measuring performance. Do you get different numbers with and
without the hal daemon running?

Thanks,
David


_______________________________________________
hal mailing list
hal at freedesktop.org
http://freedesktop.org/mailman/listinfo/hal