On Tue, Mar 16, 2010 at 7:24 AM, Pierre C <lists@xxxxxxxxxx> wrote:
I wonder about something, too : if your DB size is smaller than RAM, you could in theory set shared_buffers to a size larger than your DB provided you still have enough free RAM left for work_mem and OS writes management. How does this interact with the logic which prevents seq-scans hogging shared_buffers ?
I think the logic you are referring to is the clock sweep buffer accounting scheme. That just makes sure that the most popular pages stay in the buffers. If your entire db fits in the buffer pool then it'll all get in there real fast.
Two things to consider though:
1. The checkpoint issue still stands.
2. You should really mess around with your cost estimates if this is the case. If you make random IO cost the same as sequential IO postgres will prefer index scans over bitmap index scans and table scans which makes sense if everything is in memory.
