On Fri, 30 Apr 2004, Gary Doades wrote:
> Yes, you're right and I have done this just to prove to myself that it
> is the index scan that is the bottleneck. I have some complex SQL that
> executes very quickly with Postgres, similar to MSSQL, but the index
> scans in most of those only touch a few rows for a few loops. It seems
> to be a problem when the index scan is scanning very many rows and for
> each of these it has to go to the table just to find out if the index it
> just looked at is still valid.
>
Another way to speed this up is the TODO item: "Use bitmaps to fetch
heap pages in sequential order" For an indexscan that fetches a number
of rows those rows may be anywhere in the base table so as each index
entry is found it fetches the corresponding table row from the heap. This
is not ideal because you can be jumping around in the heap and end up
fetching the same page multiple times because table rows are in the same
page, but were found in different places in the index.
Kris Jurka