Subject | RE: [firebird-support] Query Performance - correlated subqueries and group by statements |
---|---|
Author | Aaron Abend |
Post date | 2005-04-15T16:47:31Z |
Thanks for all of the suggestions/comments.
I checked out the impact of hyperthreading and did not see much difference
in performance. Granted, the application itself suffers with HT turned off,
but the performance was so close to the original (13:07 vs 13:26 w/o ht)
that I infer that HT has little or no impact on my application (.NET and
Firebird 1.5) either way. In any case, no 100x or even 10x impact.
We have isolated the problem and determined that the application spends 3
minutes one phase, and 10 minutes on another - the 10 minute phase is our
focus, and it is all correlated subqueries.
I set DefaultDbCachePages to a high (probably ridiculous) number - 40,000.
FBServer is now using 50M, which is actually about where I want it. I
believe the page size is the default. Could changing that help?
Question: Does FB2.0 do anything new with correlated subqueries? We have
tested by manually determining inner values and found that the individual
queries are taking nanoseconds, but when the whole thing is put together as
a correlated query, it tanks.
I will post our discoveries on this, or the code, depending on what we
learn.
Thanks,
Aaron Abend
I checked out the impact of hyperthreading and did not see much difference
in performance. Granted, the application itself suffers with HT turned off,
but the performance was so close to the original (13:07 vs 13:26 w/o ht)
that I infer that HT has little or no impact on my application (.NET and
Firebird 1.5) either way. In any case, no 100x or even 10x impact.
We have isolated the problem and determined that the application spends 3
minutes one phase, and 10 minutes on another - the 10 minute phase is our
focus, and it is all correlated subqueries.
I set DefaultDbCachePages to a high (probably ridiculous) number - 40,000.
FBServer is now using 50M, which is actually about where I want it. I
believe the page size is the default. Could changing that help?
Question: Does FB2.0 do anything new with correlated subqueries? We have
tested by manually determining inner values and found that the individual
queries are taking nanoseconds, but when the whole thing is put together as
a correlated query, it tanks.
I will post our discoveries on this, or the code, depending on what we
learn.
Thanks,
Aaron Abend