Hi Matthew,

On Mon, 26 Feb 2001, Matthew A. Bacchi wrote:

>       I wonder about this comment for two reasons.  First, I have 7133 D40
> drawers which don't appear to be problematic.  But a month or so ago we
> rebuilt a server and found that "Enable Fast-Write Cache" was not set to
> "Yes".  Is it possible that you had a similar problem?  We identified it as a
> problem when the recovery log format took at least 4 times longer than we
> expected.

I do think our D40 drawer (5 x 36gb disks) was setup poorly.  I've been
over the hardware config with our IBM engineer and he thinks it's probably
not running at the full loop speed (SSA160).  We have a mixture of 40's
and 160s on the same loop which is crossed over in a HA configuration
between two serialraid 160 adaptors.  Fast-Write is disabled, I just
checked.

> Secondly, I'm in the process of doing some performance tuning on another
> system with this hardware, and find no real problems with Raid 5 on this dasd
> type.  How did you notice that AIX was waiting for the drawer?  The "monitor"
> program?

Yes, and iostat.  We have node co-location enabled on our onsite tape pool
and file caching enabled on the disk backup pool.  We had 2 migrates
running (6 DLT 7000 tape drives in a Stotek 9710) and we were only getting
4Mb/s with 100% utilisation out of the D40.  Not enough to stream even 1
drive.

> What kind of performance is the Shark advertised as doing?  What do you
> actually see with this hardware?

I defined 200Gb in 10 x 20Gb disks spread across both clusters in our
Shark E20 and then striped the backup pool logical volume across all of
them.

I just ran Bonnie against the Shark and the D40 and got this below;

              -------Sequential Output-------- ---Sequential Input--  --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block---  --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec  %CPU
ESS-E20  1024 16870 98.9 34518 42.4 10638 20.7 12025 85.2 21873 23.4  2244.8 
35.4

              -------Sequential Output-------- ---Sequential Input--  --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block---  --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec  %CPU
D40      1024 16660 96.6 41448 54.4  3436  6.8 11233 80.1 23215 25.5 476.4   6.9

Take these with a grain of salt.  I just noticed the scsi adaptor
leading to the shark had;

"Ultra speed on SE bus with external cable"

disabled.  I'm not sure if this effects the shark connection as it is
differential.


Carl.