Re: Postgres batch write very slow - what to do

От: Tom Lane
Тема: Re: Postgres batch write very slow - what to do
Дата: ,
Msg-id: 29684.1174011339@sss.pgh.pa.us
(см: обсуждение, исходный текст)
Ответ на: Re: Postgres batch write very slow - what to do  (femski)
Список: pgsql-performance

Скрыть дерево обсуждения

Postgres batch write very slow - what to do  (femski, )
 Re: Postgres batch write very slow - what to do  (Heikki Linnakangas, )
  Re: Postgres batch write very slow - what to do  (femski, )
   Re: Postgres batch write very slow - what to do  (Tom Lane, )
    Re: Postgres batch write very slow - what to do  (Heikki Linnakangas, )
     Re: Postgres batch write very slow - what to do  (Tom Lane, )
   Re: Postgres batch write very slow - what to do  (femski, )
    Re: Postgres batch write very slow - what to do  ("Merlin Moncure", )
     Re: Postgres batch write very slow - what to do  (femski, )
      Re: Postgres batch write very slow - what to do  ("Merlin Moncure", )
       Re: Postgres batch write very slow - what to do  (femski, )
        Re: Postgres batch write very slow - what to do  (Tom Lane, )
        Re: Postgres batch write very slow - what to do  ("Merlin Moncure", )
         Re: Postgres batch write very slow - what to do  ("Merlin Moncure", )
 Re: Postgres batch write very slow - what to do  ("Merlin Moncure", )
 Re: Postgres batch write very slow - what to do  ("Merlin Moncure", )
 Re: Postgres batch write very slow - what to do  ("Joshua D. Drake", )
  Re: Postgres batch write very slow - what to do  (Carlos Moreno, )
   Re: Postgres batch write very slow - what to do  ("Joshua D. Drake", )
 Re: Postgres batch write very slow - what to do  ("hubert depesz lubaczewski", )

femski <> writes:
> If 17k record/sec is right around expected then I must say I am little
> disappointed from the "most advanced open source database".

Well, the software is certainly capable of much more than that;
for instance, on a not-too-new Dell x86_64 machine:

regression=# \timing
Timing is on.
regression=# create table t1(f1 int);
CREATE TABLE
Time: 3.614 ms
regression=# insert into t1 select * from generate_series(1,1000000);
INSERT 0 1000000
Time: 3433.483 ms

which works out to something a little shy of 300K rows/sec.  Of course
the main difference from what I think you're trying to do is the lack of
any per-row round trips to the client code.  But you need to look into
where the bottleneck is, not just assume it's insoluble.

            regards, tom lane


В списке pgsql-performance по дате сообщения:

От: Alexey Romanchuk
Дата:
Сообщение: Re: Determine dead tuples size
От: Heikki Linnakangas
Дата:
Сообщение: Re: Determine dead tuples size