As a followup to my previous post, here's what a healthy, well behaved
but running under moderate load db server looks like:
top - 15:47:51 up 436 days, 2:31, 3 users, load average: 12.03, 11.86, 12.26
Tasks: 394 total, 7 running, 387 sleeping, 0 stopped, 0 zombie
Cpu(s): 17.3%us, 1.3%sy, 0.0%ni, 80.6%id, 0.4%wa, 0.1%hi, 0.3%si, 0.0%st
Mem: 33031252k total, 32623600k used, 407652k free, 282760k buffers
Swap: 9767480k total, 3768k used, 9763712k free, 21242700k cached
Note load is 12, swap is 3Meg, and IO Wait is 0.4%.
vmstat shows about 22Megs/second being written, most of that is WAL
and logs, which we can see with "iostat -x 10" which shows about 44k
512 blocks per second going to /dev/sda, my pg_xlog / OS drive set.
The single most important measurement here is IO Wait being low. If
that's low the machine may degrade slowly under heavier and heavier
load, but it won't just grind to a halt like high IO Wait will do.