How can we reach 9K IOPs for our 14 disk system?
We have SCSI IDs (disk) 0-6 on one backplane and SCSI IDs 7-13 (disk) on another backplane. Each backplane is connected to one channel of a two-channel controller. We have two RAID-1 (2 disks) each for tempdb and log file respectively. The other 10 disks are RAID 1+0 and contains indexes and tables. Moving from 10K RPM disks to 15K RPM disks (all U320 including backplane and controller) gave us increase of IOPs from 4K to 6K (total as measured from total three RAID arrays). How can we reach 9K IOPs for our 14 disk system? Is Fibre Channel the only way?
Moving to Fibre Channel would mean changing the enclosure and controllers, adding HBA and possibly new disks. Seems like this would be twice as expensive as our current solution. It might be worth if if we could get 11K IOPs for our investment. How can I measure?
There is no way that I can tell where the bottleneck is in your system.
It may be I/O bound, it may be processor bound. If it is processor bound then there is no storage solution to help you. I would put log file disk drives on a dedicated channel. Don't use RAID for logs, only use mirroring and do the mirroring in the controller, of course. This might require adding a controller for that purpose. The log files get the most activity and the most contention for resources -- separating them from tablespace storage can help. Another thing you might try is short stroking drives. In other words partition them to smaller capacity than the drive maximum -- say half the size, if you can do it. If you need all the capacity then this obviously won't work.
This was first published in July 2004