On 2011-08-11, at 18:17 , Greg Smith wrote:
> shared_buffers=512MB
> wal_buffers=16MB
> checkpoint_segments=64
Thanks for the advice. I tried these values... And it is even worse - went up to 63 minutes (from 60 minutes). Like I said this load is read mostly. My 80 / 20% might be a bit inaccurate, if so it could be more like 90% read and 10% write. I do not see any checkpoint warnings in my logs.
I guess that means the OS cache is better for this particular use case than the postgresql cache?