Re: many to many performance

Chavdar Kopoev wrote:

> I want to use as a data storage postgresql. Tried several data structures, testing btree, gin, gist indecies over them, but best achieved performance for a 10 times smaller dataset (10k cat ids, 100k doc ids, 1m relations) is slower more than 5 times.

Can you post your queries and table definitions so people trying to help
you know what you did / tried to do? A downloadable self contained
example might also be useful.

Please also post the output of `EXPLAIN' on your queries, eg:

EXPLAIN SELECT blah, ... FROM blah;

> I read about postgresql bitmap indecies and "data lookup" when scanning indecies to get a value for current transaction. Downloaded latest v8.4 snapshot, compiled it, but as I see there is no bitmap index in it. Maybe if I download HEAD revision I will find them there, dont know.

Bitmap index scans are an internal function that's used to combine two
indexes on the fly during a query (or at least use both of them in one
table scan). You don't make a bitmap index, you just make two ordinary
btree indexes and let Pg take care of this for you.

If you query on the columns of interest a lot, you might want to use a
multi-column index instead.

--
Craig Ringer