Re: select max() much slower than select min() - Mailing list pgsql-performance

Kevin Grittner [Kevin.Grittner@wicourts.gov] wrote:
> Is there any correlation between ts_id and ts_interval_start_time?
only vaguely: increasing ts_interval_start_time implies increasing ts_id
but there may be many rows (100,000's) with the same ts_interval_start_time

> Perhaps if you tried min and max with different time ranges it would
> find a row on a backward scan faster.  It'll take ten times as long if
> it has to scan through ten times as many rows to find a match.
it looks like there are fewer rows backwards than forwards:

cemdb=> select count(*) from ts_stats_transet_user_interval where
ts_interval_start_time < '2009-6-16 01:00';
   count
----------
  32100000
(1 row)

cemdb=> select count(*) from ts_stats_transet_user_interval where
ts_interval_start_time >= '2009-6-16 02:00';
   count
----------
  13500000
(1 row)


> I don't suppose you have an index on ts_interval_start_time?
there is an index. I mentioned this in my orginal posting.

Thanks,
Brian