pg.ddx.io pgsql-performance@postgresql.org mailing list archive
help / color / mirror / Atom feedGEQO Benchmark
5+ messages / 4 participants
[nested] [flat]
* GEQO Benchmark
@ 2008-05-28 13:50 tarcizioab@c3sl.ufpr.br
0 siblings, 1 reply; 5+ messages in thread
From: tarcizioab@c3sl.ufpr.br @ 2008-05-28 13:50 UTC (permalink / raw)
To: pgsql-performance
Hi all,
I'm using the TPC-H Benchmark for testing of performance in PostgreSQL.
But it is composed of only 8 tables, which is not enough to call the GEQO
algorithm.
I don't want to change any of the 22 queries provided by the TPC-H to call
the GEQO, and not lose the credibility of the TPC's tool.
Is there a specific tool for testing queries with large number of tables
(12 or more), or some a documentation of tests carried out in GEQO
Algorithm?
--
Tarcizio Bini
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: GEQO Benchmark
@ 2008-05-28 14:02 Tom Lane <tgl@sss.pgh.pa.us>
parent: tarcizioab@c3sl.ufpr.br
0 siblings, 1 reply; 5+ messages in thread
From: Tom Lane @ 2008-05-28 14:02 UTC (permalink / raw)
To: tarcizioab@c3sl.ufpr.br; +Cc: pgsql-performance
tarcizioab@c3sl.ufpr.br writes:
> I'm using the TPC-H Benchmark for testing of performance in PostgreSQL.
> But it is composed of only 8 tables, which is not enough to call the GEQO
> algorithm.
See
http://www.postgresql.org/docs/8.3/static/runtime-config-query.html#RUNTIME-CONFIG-QUERY-GEQO
particularly geqo_threshold.
regards, tom lane
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: GEQO Benchmark
@ 2008-05-28 16:13 Tarcizio Bini <tarcizioab@c3sl.ufpr.br>
parent: Tom Lane <tgl@sss.pgh.pa.us>
0 siblings, 1 reply; 5+ messages in thread
From: Tarcizio Bini @ 2008-05-28 16:13 UTC (permalink / raw)
To: pgsql-performance
Hi,
Of course, the geqo_threshold can be changed so that the geqo be performed
in queries that have less than 12 tables. However, we aim to test the GEQO
algorithm in conditions where the standard algorithm (dynamic programming)
has a high cost to calculate the query plan.
--
Tarcizio Bini
2008/5/28 Tom Lane <tgl@sss.pgh.pa.us>:
> tarcizioab@c3sl.ufpr.br writes:
> > I'm using the TPC-H Benchmark for testing of performance in PostgreSQL.
> > But it is composed of only 8 tables, which is not enough to call the GEQO
> > algorithm.
>
> See
>
> http://www.postgresql.org/docs/8.3/static/runtime-config-query.html#RUNTIME-CONFIG-QUERY-GEQO
>
> particularly geqo_threshold.
>
> regards, tom lane
>
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: GEQO Benchmark
@ 2008-05-28 16:52 Simon Riggs <simon@2ndquadrant.com>
parent: Tarcizio Bini <tarcizioab@c3sl.ufpr.br>
0 siblings, 1 reply; 5+ messages in thread
From: Simon Riggs @ 2008-05-28 16:52 UTC (permalink / raw)
To: Tarcizio Bini <tarcizioab@c3sl.ufpr.br>; +Cc: pgsql-performance
On Wed, 2008-05-28 at 13:13 -0300, Tarcizio Bini wrote:
> Of course, the geqo_threshold can be changed so that the geqo be
> performed in queries that have less than 12 tables. However, we aim to
> test the GEQO algorithm in conditions where the standard algorithm
> (dynamic programming) has a high cost to calculate the query plan.
My understanding is the GEQO cannot arrive at a better plan than the
standard optimizer, so unless you wish to measure planning time there
isn't much to test. What is the quality of the plans it generates? Well
that varies according to the input; sometimes it gets the right plan,
other times it doesn't get close.
--
Simon Riggs www.2ndQuadrant.com
PostgreSQL Training, Services and Support
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: GEQO Benchmark
@ 2008-05-28 17:54 Tarcizio Bini <tarcizioab@c3sl.ufpr.br>
parent: Simon Riggs <simon@2ndquadrant.com>
0 siblings, 0 replies; 5+ messages in thread
From: Tarcizio Bini @ 2008-05-28 17:54 UTC (permalink / raw)
To: pgsql-performance
I'm interested in the response time instead of the quality of it, that is,
if this response is (or not) a best plan. Thus, my interest on the response
time is to compare the GEQO algorithm with other solutions for optimization
of queries, more specificaly when there are 12 or more tables evolved.
Accordingly, the algorithm of dynamic programming becomes hard because of
the computational cost to obtain the optimal plan.
My question is to know what tool or method was used to test the GEQO
algorithm or if there is some documentation that explain how the geqo was
tested.
Regards,
Tarcizio Bini
2008/5/28 Simon Riggs <simon@2ndquadrant.com>:
>
> On Wed, 2008-05-28 at 13:13 -0300, Tarcizio Bini wrote:
>
> > Of course, the geqo_threshold can be changed so that the geqo be
> > performed in queries that have less than 12 tables. However, we aim to
> > test the GEQO algorithm in conditions where the standard algorithm
> > (dynamic programming) has a high cost to calculate the query plan.
>
> My understanding is the GEQO cannot arrive at a better plan than the
> standard optimizer, so unless you wish to measure planning time there
> isn't much to test. What is the quality of the plans it generates? Well
> that varies according to the input; sometimes it gets the right plan,
> other times it doesn't get close.
>
> --
> Simon Riggs www.2ndQuadrant.com
> PostgreSQL Training, Services and Support
>
>
^ permalink raw reply [nested|flat] 5+ messages in thread
end of thread, other threads:[~2008-05-28 17:54 UTC | newest]
Thread overview: 5+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2008-05-28 13:50 GEQO Benchmark tarcizioab@c3sl.ufpr.br
2008-05-28 14:02 ` Tom Lane <tgl@sss.pgh.pa.us>
2008-05-28 16:13 ` Tarcizio Bini <tarcizioab@c3sl.ufpr.br>
2008-05-28 16:52 ` Simon Riggs <simon@2ndquadrant.com>
2008-05-28 17:54 ` Tarcizio Bini <tarcizioab@c3sl.ufpr.br>
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox