Kurt Overberg wrote:
> work_mem = 100MB # when I ran the original query,
> this was set to 1MB, increased on Mark Kirkwood's advice, seemed to help
> a bit but not really
>
For future reference, be careful with this parameter, as *every*
connection will use this much memory for each sort or hash (i.e it's not
shared and can be allocated several times by each connection!)...yeah, I
know I suggested increasing it to see what effect it would have :-).
And I'd agree with Steiner and others, looks like caching effects are
the cause of the timing difference between production and the mac!
Cheers
Mark
