On Fri, 6 Sep 2002, Mihai Gheorghiu wrote:
> PG7.1.3 on RH7.1 on Dell PowerEdge 2500SC P3-933, 1GB RAM, 18GB SCSI160
> There are 350k rows with trxtype=MP
> With indexscan=off:
> Aggregate (cost=22975.15..22990.27 rows=302 width=24)
> -> Group (cost=22975.15..22982.71 rows=3025 width=24)
> -> Sort (cost=22975.15..22975.15 rows=3025 width=24)
> -> Seq Scan on tbas_transactions (cost=0.00..22800.29
> rows=3025 width=24)
> Time: 25.9s
> With indexscan=on:
> Aggregate (cost=10874.64..10889.76 rows=302 width=24)
> -> Group (cost=10874.64..10882.20 rows=3025 width=24)
> -> Sort (cost=10874.64..10874.64 rows=3025 width=24)
> -> Index Scan using trx_trxtype_idx on tbas_transactions
> (cost=0.00..10699.78 rows=3025 width=24)
> Time: 24.9s
> The point is I need to run this query in a fraction of the above time,
> otherwise I'm in deep trouble.
> Any suggestion is welcome.
Hmm, in general a multicolumn index on (trxtype,account) might enable it
to get rid of the sort step.