Josh Berkus <josh(at)agliodbs(dot)com> writes:
> Forget hyperthreading. Look at their postgresql.conf settings. 8mb shared
> mem, 16mb sort mem per connection for 512 connections, default
> effective_cache_size.
They could well be going into swap hell due to the oversized sort_mem,
but that didn't quite seem to explain the reported behavior. I'd want
to see vmstat or similar output to confirm whether the disks are busy,
though. Amazing how many people forget that a database is normally
I/O-bound rather than CPU-bound.
regards, tom lane