Postgres Pro
Downloads
Home   >   mailing lists

Thread: PG Statistics

PG Statistics

From
"mcelroy, tim"
Date:

Good evening,

Does anyone know how much of a performance hit turning stats_block_level and  stats_row_level on will incur?  Do both need to be on to gather cache related statistics?  I know the annotated_conf_80 document states to only turn them on for debug but if they're not that performance intensive I cannot see the harm.

Thank you,
Tim McElroy

Re: PG Statistics

From
Michael Fuhr
Date:
On Mon, Mar 13, 2006 at 06:49:39PM -0500, mcelroy, tim wrote:
> Does anyone know how much of a performance hit turning stats_block_level and
> stats_row_level on will incur?  Do both need to be on to gather cache
> related statistics?  I know the annotated_conf_80 document states to only
> turn them on for debug but if they're not that performance intensive I
> cannot see the harm.

I ran some tests a few months ago and found that stats_command_string
had a significant impact, whereas stats_block_level and stats_row_level
were almost negligible.  Here are my test results:

http://archives.postgresql.org/pgsql-performance/2005-12/msg00307.php

Your results may vary.  If you see substantially different results
then please post the particulars.

--
Michael Fuhr

Re: PG Statistics

From
Steve Poe
Date:
Tim,

When I have done ODBC load tests with stats_block_level enabled on (20
mins. per test), I've seen about 3-4% performance hit. Your mileage may
vary.

Steve Poe

On Mon, 2006-03-13 at 18:49 -0500, mcelroy, tim wrote:
> Good evening,
>
> Does anyone know how much of a performance hit turning
> stats_block_level and  stats_row_level on will incur?  Do both need to
> be on to gather cache related statistics?  I know the
> annotated_conf_80 document states to only turn them on for debug but
> if they're not that performance intensive I cannot see the harm.
>
> Thank you,
> Tim McElroy
>

Re: PG Statistics

From
"mcelroy, tim"
Date:

Thank you for the insight Michael.  I'll be performing some tests with the various setting on/off this week and will post the results.

Tim

 -----Original Message-----
From:   Michael Fuhr [mailto:mike@fuhr.org]
Sent:   Monday, March 13, 2006 7:19 PM
To:     mcelroy, tim
Cc:     'pgsql-performance@postgresql.org'
Subject:        Re: [PERFORM] PG Statistics

On Mon, Mar 13, 2006 at 06:49:39PM -0500, mcelroy, tim wrote:
> Does anyone know how much of a performance hit turning stats_block_level and
> stats_row_level on will incur?  Do both need to be on to gather cache
> related statistics?  I know the annotated_conf_80 document states to only
> turn them on for debug but if they're not that performance intensive I
> cannot see the harm.

I ran some tests a few months ago and found that stats_command_string
had a significant impact, whereas stats_block_level and stats_row_level
were almost negligible.  Here are my test results:

http://archives.postgresql.org/pgsql-performance/2005-12/msg00307.php

Your results may vary.  If you see substantially different results
then please post the particulars.

--
Michael Fuhr

Re: PG Statistics

From
"mcelroy, tim"
Date:

Thanks you Steve.  As mentioned in my other reply to Michael Fuhr I'll post the results from tests to be performed this week.

Tim

 -----Original Message-----
From:   Steve Poe [mailto:spoe@sfnet.cc]
Sent:   Monday, March 13, 2006 7:38 PM
To:     mcelroy, tim
Cc:     'pgsql-performance@postgresql.org'
Subject:        Re: [PERFORM] PG Statistics

Tim,

When I have done ODBC load tests with stats_block_level enabled on (20
mins. per test), I've seen about 3-4% performance hit. Your mileage may
vary.

Steve Poe

On Mon, 2006-03-13 at 18:49 -0500, mcelroy, tim wrote:
> Good evening,
>
> Does anyone know how much of a performance hit turning
> stats_block_level and  stats_row_level on will incur?  Do both need to
> be on to gather cache related statistics?  I know the
> annotated_conf_80 document states to only turn them on for debug but
> if they're not that performance intensive I cannot see the harm.
>
> Thank you,
> Tim McElroy
>