Re: Online 7.12 Tunning [more info]

Serafim Fonseca wrote: > snip.... Serafim - I'd start by bumping up BUFFERS to 10000 or 15000 or more until paging / swapping kicks in. With 2 disks, you can't afford much paging/swapping. If you can reduce SHMVIRTSIZE, do it (you probably can...). This will improve your read hit ratio, and reduce disk I/O. See how much this helps. If this is not enough, it looks like you need to tune some of the application SQL. I assume you have done "update statistics" lately, and this is not the cause of too many sequential scans.... Quoting from your post: ixda-RA idx-RA da-RA RA-pgsused lchwaits 423353 3758 4201547 4626920 222 da-RA is way too high, and ixda-RA and idx-RA are too low. Try to reverse this ratio. A few well placed indexes to help the most offensive SQL should reduce da-RA and increase ixda-RA, which you want. You're doing *way* too much I/O on sequential scans, and beating your 2 disks to death. vmstat wait time should be very low with well functioning I/O (no more than a few percent...). Read up on "onstat -g sql <sessionid>", and "set explain on". If this isn't enough, you'll probably have to belly up more more disks, more memory or both. Greg