We have 3 mirrored pairs of disks in our subsystem that were systematically running at between 3 and 5 times more busy than all the others. To alleviate this we ran a trace (TRCASPBAL SET(*ON) TIMLMT(1440)) over a 24 hour period, then immediately ran an ASP balance (STRASPBAL TYPE(*USAGE) TIMLMT(*NOMAX)) to move the more frequently accessed data to the less busy disks. This worked exactly as expected, however, we then had the same 3 mirrored pairs of disks running at 30% less capacity than the others in the subsystem and subsequently, the OS favoured writing to them in order to balance out the '% used' and so they are still running 3-5 times more busy than the others due to the excessive writes.
move forward 3 months
The disk subsystem has now balanced its capacity but as those 3 disks mentioned before have been written to excessively for the past 3 months... they're still running at around 40% busy with the rest of the disks in the subsystem running around 2-9% busy.
I can't see how we would ever get out of this viscous circle...
Running V5R4M0 on a partitioned i570 (2.6 processors assigned) with 21 mirrored pairs of 141Gb disks in a single ASP.
Any advice or suggestions would be much appreciated as I have performance issues on any batch process that does lots of DASD I/O.
Regards,
Tony.
move forward 3 months
The disk subsystem has now balanced its capacity but as those 3 disks mentioned before have been written to excessively for the past 3 months... they're still running at around 40% busy with the rest of the disks in the subsystem running around 2-9% busy.
I can't see how we would ever get out of this viscous circle...
Running V5R4M0 on a partitioned i570 (2.6 processors assigned) with 21 mirrored pairs of 141Gb disks in a single ASP.
Any advice or suggestions would be much appreciated as I have performance issues on any batch process that does lots of DASD I/O.
Regards,
Tony.