Обсуждение: Create Index Performance Issue

Hi,

I'm currently testing a 25G Postgres 8.3.0 database on Apple Xserve/
Intel with Mac OSX 10.5 (Leopard).

There are two identical machines with identical configurations (hw
and sw).
Running a pg_restore of a binary backup file (3.8 GB) on both
machines gives the following results:

   Machine A :  90 minutes
   Machine B:   60 minutes

The postgres server logfiles show the cause of the time difference:
The SQL query
    CREATE INDEX login_session_identifier ON login_session USING
btree (identifier)
consumes very different execution times:

   Machine A :  1905 seconds
   Machine B:     292 seconds

The difference of 27 minutes explains the pg_restore behaviour.
The same effect can be demonstrated easily by just running the SQL
query manually.

Some database infos:
   Table login_session :  33,996,225 rows,  10.6 GB on disk
    Column identifier :   character varying (255)
    Index login_session_identifier :  1.3 GB on disk

Postgres configuration file is of course the same on both machines.
The maintenance_work_mem is set to  128MB . I played with this
parameter setting it
up to 2 GB. This didn't change the execution times essentially. With
large values
the times went a little up.

I would appreciate any advices on how to investigate this problem any
further.
Ideas are very welcome.

Thanks in advance,
Rudolf VanderLeeden
Logic United GmbH, Germany
vanderleeden@logicunited.com

On Wed, Jun 04, 2008 at 11:43:24AM +0200, Rudolf van der Leeden wrote:

> I'm currently testing a 25G Postgres 8.3.0 database on Apple Xserve/
> Intel with Mac OSX 10.5 (Leopard).
>
> There are two identical machines with identical configurations (hw
> and sw).
> Running a pg_restore of a binary backup file (3.8 GB) on both
> machines gives the following results:
>
>   Machine A :  90 minutes
>   Machine B:   60 minutes

Maybe there are some background processes running?

Tino.

--
"What we resist, persists." (Zen saying)

www.craniosacralzentrum.de
www.forteego.de

Rudolf van der Leeden <vanderleeden@logicunited.com> writes:
> [ different times to build a varchar index on allegedly identical
>   installations ]

Perhaps one is using C locale and the other is not?  strcmp() vs
strcoll() is a pretty big hit.  In general it'd be worth dumping
out the whole contents of pg_settings and diffing those two files
to see if you missed any configuration differences.

            regards, tom lane

Tom,
your hint was 100% correct.  BINGO!
The setting of  lc_locale and lc_ctype was 'en_us' instead of 'C' as
on the other machine.
Now it works perfectly OK and with identical execution times on both
machines.
Thanks for your help and best regards,
Rudolf VanderLeeden

Am 04.06.2008 um 16:45 schrieb Tom Lane:

> Perhaps one is using C locale and the other is not?  strcmp() vs
> strcoll() is a pretty big hit.  In general it'd be worth dumping
> out the whole contents of pg_settings and diffing those two files
> to see if you missed any configuration differences.