[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Sucky performance under disk load



David Z Maze wrote:
This is with my Dell Latitude C600.  Somewhat recently, I've noticed
that, when there's heavy disk activity, the entire system grinds
almost to a halt.  I suspect I've noticed it more since I've installed
postgresql and imported a reasonably large data set into it; running
VACUUM FULL ANALYZE over the database will kill the system.  The daily
updatedb process also hurts a lot.

The symptom is just that everything is reeeallly
slloooowwwww... playing Vorbis files under xmms breaks a lot, the
mouse is unresponsive and jerky, and so on.  If I watch what's going
on using, say, mgm, there's somewhat minimal disk traffic (perhaps
4-8k sectors/second read, peaks up to 32k sectors) and basically no
CPU cycles being used.  Poking with top suggests that both postmaster
(the postgresql server) and kjournald are blocking; is ext3fs hurting
me unnecessarily here?

AFAIK I've set up everything that wants to be with hdparm, including
IRQ unmasking, 32-bit I/O, and DMA.  There's no particular hints in
the system logs that anything is amiss ("PIIX4: not 100% native mode:
will probe irqs later", but this doesn't sound like the end of the
world).

Any hints?

I experienced the same problem on my Dell I8100.

I now use kernel 2.4.20 with O(1) scheduler, preemptible, low latency
patches available on Con Kolivas Kernel patch homepage
(http://members.optusnet.com.au/ckolivas/kernel/).
I also added 'noatime' to /etc/fstab (in the mount man page it seemed
not that useful to me).
"/dev/hda4 / ext3 defaults,noatime,errors=remount-ro 0 1".

Since then xmms plays always smoothly during long find processes :-)

--
DoM



Reply to: