Re: qtemp performance oddity.

[DrFranken <midrange@xxxxxxxxxxxx>]

Currently there are 8 17s and 7 8s in there. Yes this can cause 
performance issues but not when percent busy on the drives is almost 
always single digits. Issue is that with twice the data on the 17s as 
the 8s they would in theory be twice as busy. In fact they are about 
twice as busy, 8% vs 4%.

Storage was balanced after disk work was complete. Righ now it's spon on 
53.1% on all 15 drives.

RCLSTG, Hmmmmmm. Possibility.


        - Larry "DrFranken" Bolhuis

www.frankeni.com
www.iDevCloud.com
www.iInTheCloud.com

On 2/27/2013 10:30 PM, Roger Harman wrote:

> A couple of SWAG's....
>
> Are there still some 8G drives in there?  On same controller?  I was told a
> long time ago that that would cause a big performance hit.
>
> Is storage balanced across the drives?
>
> RCLSTG, perhaps?
>
>
>
> -----Original Message-----
> From: midrange-l-bounces@xxxxxxxxxxxx
> [mailto:midrange-l-bounces@xxxxxxxxxxxx] On Behalf Of DrFranken
> Sent: Wednesday, February 27, 2013 7:21 PM
> To: Midrange Systems Technical Discussion
> Subject: qtemp performance oddity.
>
> Customer with an old 720 4-way, old enough to be running V5R1.
>
> A couple weeks ago we replaced 8 7200 RPM 8G drives with 10K RPM 17GB
> Drives. Also added 768MB Memory lifting them to 2G and replaced 2741 RAID
> card with 4778. Disks were drained/added-no save restore. PTFs are VERY
> close to the latest available for V5R1.
>
>
> System had run for nearly 2 years without IPL before the upgrade.
>
> Since the 'upgrade' the system is slower. CPU usage usually in single
> digits, paging and faulting numbers are low double digits, disk percent
> busys might hit 10 (15 total drives all 10K and 52% full.) WRKSYSACT and
> performance tools reports show nothing odd at all. Seizes and locks are low,
> disk response times are very good. IOP and IOA processor utilizations are
> 10% or less. cfint shows 0.0 or 0.1% in WRKSYSACT.
>
> Nothing at all in lic logs, problem logs, etc and yes Cache battery is good.
> :-)
>
> Any job that puts any amount of stuff in QTEMP just dogs. Some that used to
> take 30 seconds take 10 minutes. Users who run tasks that use QTEMP see more
> than two minutes to SIGNOFF.
>
> I did some testing. Create file with 1M 1K Records (RPG) took 9 1/2 minutes
> (==1750 records per second and pushed CPU to 20%, disks to 25%
> busy) Copying said file to another lib 4 1/2 minutes, copying to QTEMP 4
> 1/2 minutes. Clearing the other lib 30 seconds, clearing QTEMP 30 seconds.
> Signoff with empty QTEMP-immediate, Signoff with that file in
> QTEMP-immediate. This tells me that basic stuff works fine.
>
> Software in use is old JBA System-21. No source available but then no
> changes were made.
>
> Any ideas? Anyone recognize these symptoms?
>
> This is odd.
>