Re: Yet another vectorized engine
От | Konstantin Knizhnik |
---|---|
Тема | Re: Yet another vectorized engine |
Дата | |
Msg-id | 7381c2ee-9776-9cff-dd7e-23345bcba6d7@postgrespro.ru обсуждение исходный текст |
Ответ на | Re: Yet another vectorized engine (Hubert Zhang <hzhang@pivotal.io>) |
Ответы |
Re: Yet another vectorized engine
Re: Yet another vectorized engine |
Список | pgsql-hackers |
I have done some performance comparisons.
First of all I failed to run vectorized version of Q1 with master branch of your repository and
PG9_6_STABLE branch of Postgres:
NOTICE: query can't be vectorized
DETAIL: Non plain agg is not supported
I have to switch to pg96 branch.
Results (seconds) of Q1 execution are the following:
I failed to run parallel version of Q1 with enable_vectorize_engine=on because of the same error: "Non plain agg is not supported"
So looks like PG-13 provides significant advantages in OLAP queries comparing with 9.6!
Definitely it doesn't mean that vectorized executor is not needed for new version of Postgres.
Once been ported, I expect that it should provide comparable improvement of performance.
But in any case I think that vectorized executor makes sense only been combine with columnar store.
First of all I failed to run vectorized version of Q1 with master branch of your repository and
PG9_6_STABLE branch of Postgres:
NOTICE: query can't be vectorized
DETAIL: Non plain agg is not supported
I have to switch to pg96 branch.
Results (seconds) of Q1 execution are the following:
max_parallel_workers_per_gather | PG9_6, enable_vectorize_engine=off | PG9_6, enable_vectorize_engine=on | master (jit=on) |
0 | 36 | 20 | 10 |
4 | 10 | - | 5 |
I failed to run parallel version of Q1 with enable_vectorize_engine=on because of the same error: "Non plain agg is not supported"
So looks like PG-13 provides significant advantages in OLAP queries comparing with 9.6!
Definitely it doesn't mean that vectorized executor is not needed for new version of Postgres.
Once been ported, I expect that it should provide comparable improvement of performance.
But in any case I think that vectorized executor makes sense only been combine with columnar store.
-- Konstantin Knizhnik Postgres Professional: http://www.postgrespro.com The Russian Postgres Company
В списке pgsql-hackers по дате отправления: