On Thursday, Jul 1, 2004, at 02:52 Australia/Sydney, Stef Coene wrote:

On Tuesday 29 June 2004 20:33, asr AT UFL DOT EDU wrote:

==> In article <200406291330.i5TDUEmp001568 AT igate.highmark DOT com>, Joni
Moyer

<joni.moyer AT HIGHMARK DOT COM> writes:

Hello all!

I was reading the performance tuning guide and it states that we
should
use raw partitions for server db, log and disk storage pool volumes
for
an AIX server and I was just wondering if this is true and what the
benefits are of configuring volumes in this manner?

Simpler, faster, less space overhead.

Euh, yes and no.  For AIX and jfs2 file systems, you can enable CIO
(in /etc/filesystems: "options = rw,cio").  If you do so, your file
systems
are as fast as raw devices.  So you have the benefits of a file system
and
the speed of a raw device.  The I/O requests are directly done on the
disk,
all cache is skipped.
I have a pdf file about this setup for oracle and the speed you can
get.  We
once enabled this on a very busy AIX server and the oracle database
was very,
very fast.

Not quite true. From my research, TSM imposes its own per-volume I/O
serialisation, so will not benefit at all from Concurrent I/O.
However, it will benefit from Direct I/O (mount option "dio") which
allows I/O to bypass the buffer cache.

On TSM 5.1 and up, the "AIXDIRECTIO" option was introduced, and is
enabled by default. However, the following note appears in RedBook
SG24-6554-00:

Note: Please note that currently direct I/O is only used for disk
storagepool volumes, not
database or log volumes, and only on non-compressed and
non-large-file-enabled file
systems. It is also only supported for up to 2 GB volumes.

My preference is still raw logical volumes - the fact that volume
formatting is not required is just one small reason. Then, configure
TSM to use asynchronous I/O with the AIXASYNCIO dsmserv.opt option.

Cheers,
--
Paul Ripke
Unix/OpenVMS/TSM/DBA
I love deadlines. I like the whooshing sound they make as they fly by.
-- Douglas Adams