Re: [PERFORM] How does PG know if data is in memory?

2010-10-01 Thread Fabrício dos Anjos Silva
ry different parameters, run a load test, process results and generate reports? Again, thanks all of you for the replies. Cheers, Fabrício dos Anjos Silva LinkCom Soluções em T.I. 2010/10/1 Kevin Grittner > Craig Ringer wrote: > > > Because the query often only wants

Re: [PERFORM] How does PG know if data is in memory?

2010-10-01 Thread Fabrício dos Anjos Silva
ying to get the best of it. Unfortunately, it does not seem to be an easy task. If someone could point good books about PG tuning, I would appreciate that. I found some yet to be released books about PG 9. Any comments about them? Thank you all. Fabrício dos Anjos Silva LinkCom Soluções em T

[PERFORM] Wrong index choice

2010-09-29 Thread Fabrício dos Anjos Silva
ter the first returned value. The estimated Limit cost of 0.00 is ok, even if the Scan cost is estimated at 42146.33. I only managed to get all data in OS cache after mounting this new server with 7,7GB of memory, which I can't afford to use permanently. My real server has only 1,7GB of memory and this Index Scan Backward plan takes forever to run (I really don't know how much time), making tons of random seeks. When enable_indexscan id off, the query runs quickly on the 1,7GB server. Initially I was using statistics for these 2 columns as 200 and 300, but even after changing to 1000, the problem persists. I tried several different values the seq_page_cost, random_page_cost, cpu_tuple_cost, cpu_index_tuple_cost and cpu_operator_cost with no success. Could someone explain me this? Thanks, Fabrício dos Anjos Silva LinkCom Soluções em T.I.

[PERFORM] How does PG know if data is in memory?

2010-09-29 Thread Fabrício dos Anjos Silva
disk" and random_page_cost mean "random access on disk"? I appreciate if someone could clear this out. Thanks! Fabrício dos Anjos Silva LinkCom Soluções em T.I.