On Thu, Oct 23, 2008 at 8:48 PM, Scott Carey <scott@richrelevance.com> wrote:
> If you are doing batch inserts of data, and want to have reporting queries
> concurrently running, make sure you have the pg_xlogs on a different disk
> than the data/indexes. 2 drives RAID 1 for OS + xlogs works great (and
From the OPs original post I'd guess that one big RAID 10 would serve
him best, but yeah, you need to test to really see.
> Also, if you intend to have lots of data organized by a time field, and
> expect to do the reporting/aggregation queries on subsets of that data
> bounded by time, partitioning by time can have huge benefits. Partition by
> month, for example, and sequential scans will only flow to the months of
> interest if the queries have the right lmits on the date in the where
> clause.
I second this. Partitioning in time in past reporting databases
resulted in huge performance improvements for select queries.