On Mon, 01 Oct 2007, Mike Charnoky wrote: > I altered the table in question, with "set statistics 100" on the > timestamp column, then ran analyze. This seemed to help somewhat. Now, > queries don't seem to hang, but it still takes a long time to do the count: > * "where evtime between '2007-09-26' and '2007-09-27'" > took 37 minutes to run (result was ~12 million) > * "where evtime between '2007-09-25' and '2007-09-26'" > took 40 minutes to run (result was ~14 million) Maybe it needs several million scattered seeks which basically disable disk cache. If you can afford a database inactivity period you can for example do periodically (for example on nights or weekends): cluster pred_acc_evtime_index on prediction_accuracy; (see http://www.postgresql.org/docs/8.2/static/sql-cluster.html) This would run rather long time for the first time, but will be much faster later. This should make seeks much more local and it would help operating system to cache results. Regards Tometzky -- ...although Eating Honey was a very good thing to do, there was a moment just before you began to eat it which was better than when you were... Winnie the Pooh ---------------------------(end of broadcast)--------------------------- TIP 6: explain analyze is your friend
