[Linux-cluster] gfs performance

[Gabe Turner]

On Mon, Jan 30, 2006 at 03:46:27PM +0100, Jure Pe??ar wrote:
> On Mon, 30 Jan 2006 07:12:51 -0600
> Eric Anderson <anderson at centtech.com> wrote:
> 
> > I'm not sure what kind of disk backend you have, but you are probably 
> > thrashing the heads on the disks, making the disks do a lot of seeks, 
> > burning your performance.  I don't think this is a GFS performance issue 
> > really.  You could try a memory backed disk with GFS on it, and see if 
> > the performance falls at all.
> > 
> I wrote that down under comments:
> "Shared storage is IBM DS4100, configured with two raid5 volumes of 7 disks each"
> 
> It's true that DS4100 is no high performer, but I'd say all that write cache should make a difference with seeks?
> 

The DS4100, provided your's has two controllers, should have about 512MB of
RAID cache total.  If this is the LSI/Engenio-based SATA storage I'm
thinking of, you should be able to turn on write cache on the controllers
via the management interface while the unit is live in order to see if it
makes much of a difference.  I manage a number of Engenio-based RAIDs
(including some SATA) and enabling write cache gave me a 2-4x increase in
performance when doing small I/O (e.g. using the SATA as the storage pool
for a large tape library).

HTH,

Gabe
-- 
Gabe Turner                                             gabe at msi.umn.edu
UNIX Systems Administrator,
University of Minnesota Supercomputing Institute
 for Digital Simulation and Advanced Computation         www.msi.umn.edu