On Thu, May 31, 2012 at 11:09 AM, Sergey Koposov <koposov@ast.cam.ac.uk> wrote:
> On Thu, 31 May 2012, Simon Riggs wrote:
>
>>
>> That struck me as a safe and easy optimisation. This was a problem I'd
>> been trying to optimise for 9.2, so I've written a patch that appears
>> simple and clean enough to be applied directly.
>
>
> Thanks! The patch indeed improved the timings, The dropping of 100 tables in
> a single commit before the patch took ~ 50 seconds, now it takes ~ 5 sec (it
> would be nice to reduce it further though, because the dropping of 10000
> tables still takes ~10 min).
I'm surprised it helped that much. I thought the most it could
theoretically could help would be a factor of 4.
I tried the initially unlocked test, and for me it cut the time by a
factor of 3. But I only have a 1GB shared_buffers at the max, I would
expect it help more at larger sizes because there is a constant
overhead not related to scanning the shared buffers which gets diluted
out the larger shared_buffers is.
I added to that a drop-all very similar to what Simon posted and got
another factor of 3.
But, if you can do this during a maintenance window, then just
restarting with a much smaller shared_buffers should give you a much
larger speed up than either or both of these. If I can extrapolate up
to 10G from my current curve, setting it to 8MB instead would give a
speed up of nearly 400 fold.
Cheers,
Jeff