On Nov 12, 2007 11:03 PM, Mason Hale <masonhale@xxxxxxxxx> wrote:
> > your i/o must be really random to be seeing numbers that lousy (10
> > seconds to replay a file is 1.6 megabytes/sec), or there is some other
> > unexplained problem with your server. is your raid controller
> > properly caching wites? have you benchmarked the volume with bonnie++
> > or similar tool (pay close attention to seeks).
> Here's the bonnie++ output (two runs):
> Version 1.03 ------Sequential Output------ --Sequential Input-
> --Random-
> -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
> --Seeks--
> Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec
> %CP
> dev-db-2 32G 43174 99 87421 24 45614 12 48302 97 164574 23 205.3
your seeks are low, especially if these are 15k drives. The
sequential numbers are ok but not very important in the scheme of
things. Interestingly, your 'sdd' device is not doing any writing in
the iostat samples you sent...is that a typical sample? how long are
your iostat intervals? your iowait numbers are also remarkably
stable. did you iostat the device when doing bonnie? (an iostat
during wal replay is much more interesting)
here are some random suggestions:
* play with partial wal writes setting and see if that helps
* double check raid controller is configured for writeback (it should,
with a bbu)
* experiment with xfs on data volume which may help compensate for
lousy seeking hardware
* try and describe with a little more detail your workload on the primary
merlin
---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
choose an index scan if your joining column's datatypes do not
match
