Are there any downsides to setting
the AIX queue_depth for SAN hdisks
to high values (e.g 24 ) for DB2
databases?

At my current site, the storage is Hitachi
dual fibre attached 99xx, and the database
server box and the SAN space is dedicated
to the application.

I'm seeing poor average page-read/page-write
times for a busy 32-bit DB2 v7.x database
which uses single-container File-DMS tablespaces
on a SAN all of which are accessible via a
single hdisk pair whose queue_depth is
mysteriously left at 2.


I want to increase queue_depth up to 24 (from a maximum of 32) to see
if I can reduce the I/O waits and get better average
page read/write times.

The database is overwhelming doing SELECT rather than
insert/delete/update actions, and bufferpools are
maxed-out for 32-bit memory, with 99% or more
hit-ratios.

The admin folks for the SAN confirm that
the DB2 box is not stressing the SAN, which
suggests to me that the bottleneck is at the
AIX end, not at the SAN end.


Some facts:
(DB2_PARALLEL_IO=*, DB2_STRIPED_CONTAINERS is set,
the ratio of PREFETCHSIZE:EXTENTSIZE is 4:1 ,
and NUM_IOSERVERS=64 and NUM_IOCLEANERS=32 ).


We plan to test with higher queue_depths
but I am asking if anyone else has any
stories about the impact of increasing
the queue_depth, and in particular
whether there are any possible negative
impacts or potential-risks associated with increased
queue_depth.