Raymond Burgess wrote:
I wouldn't expect sys% to be up with a scsi disk sub system. The system is running software raid, but raid5d isn't using a lot of cputime either
I have run vmstat.
'vmstat 5 50' => http://hostility.deeper.co.nz/~raymond/vmstat.out
There is often 1 proc listed as blocked so I guess the harddrives just can't keep up.
20% system is "high".
When doing mass imports, it's much better to run with no indexes, import the entire data, then regenerate the indexes at the end. Also, make sure if you're running a recent postgres that the autovacuum daemon isn't running revacuuming your data regularly. As Oliver mentioned, using COPY instead of INSERT makes a significant difference (I think it does the above itself, instead of having to do it manually).