Prev: Remaining Streaming Replication Open Items
Next: pgsql: Check compulsory parameters in recovery.conf in standby_mode, per
From: "David E. Wheeler" on 6 Apr 2010 14:06
On Apr 6, 2010, at 10:52 AM, David E. Wheeler wrote:

> Oh FFS! I was looking at the wrong build script. It was indeed built with --enable-cassert --enable-debug. Grrr.
>
> Well, that's likely the culprit right there. I'm rebuilding without those now and hopefully my tests will be back down to 45s.
>
> Many apologies for the noise and wasted time.

And just to close out this thread, I rebuilt without `--enable-cassert --enable-debug` and now the tests pass in 57s and 46s over two runs, just like on 8.4 (though with just the default postgresql.conf, unlike my 8.4 install).

Phew! Knew it had to be somehting stup^H^H^Himple.

Best,

David


--
Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

From: Tom Lane on 6 Apr 2010 14:15
"David E. Wheeler" <david(a)kineticode.com> writes:
> On Apr 6, 2010, at 10:17 AM, Tom Lane wrote:
>> So I'm not sure where your 15x is coming from, but I don't see it.

> By stupidly having configured with --enable-cassert --enable-debug without realizing it. I've just rebuilt without them and run the tests again using the default postgresql.conf and I'm back down to 57s and 46s over two runs.

Huh. I'm still curious, because in my test the overhead of those
options seemed to be about 3x. So there's still something considerably
different between what you did and what I did.

Are you testing a separate connection per ping query? I think I
neglected to compare that case with and without assert overhead.

regards, tom lane

--
Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

From: Greg Smith on 6 Apr 2010 14:28
David E. Wheeler wrote:
> By stupidly having configured with --enable-cassert --enable-debug without realizing it. I've just rebuilt without them and run the tests again using the default postgresql.conf and I'm back down to 57s and 46s over two runs.
>

Every performance test I run, regardless of where the binaries come from
or how I thought they were built, starts like this:

postgres=# show debug_assertions;
debug_assertions
------------------
off
(1 row)


It's a really good habit to get into, or even enforce in your testing
script if practical.

--
Greg Smith 2ndQuadrant US Baltimore, MD
PostgreSQL Training, Services and Support
greg(a)2ndQuadrant.com www.2ndQuadrant.us


--
Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

From: "David E. Wheeler" on 6 Apr 2010 14:39
On Apr 6, 2010, at 11:15 AM, Tom Lane wrote:

>> By stupidly having configured with --enable-cassert --enable-debug without realizing it. I've just rebuilt without them and run the tests again using the default postgresql.conf and I'm back down to 57s and 46s over two runs.
>
> Huh. I'm still curious, because in my test the overhead of those
> options seemed to be about 3x. So there's still something considerably
> different between what you did and what I did.

Are you doing this on a Mac?

> Are you testing a separate connection per ping query? I think I
> neglected to compare that case with and without assert overhead.

No, should be one connection for the entire test suite.

Best,

David
--
Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

From: Tom Lane on 6 Apr 2010 15:50
"David E. Wheeler" <david(a)kineticode.com> writes:
> On Apr 6, 2010, at 11:15 AM, Tom Lane wrote:
>> Huh. I'm still curious, because in my test the overhead of those
>> options seemed to be about 3x. So there's still something considerably
>> different between what you did and what I did.

> Are you doing this on a Mac?

I hadn't, but since you mention it: 100000 iterations take about 7.5sec
with non-assert CVS HEAD and 15sec with asserts, on a 2008 Macbook Pro.
Color me still confused.

regards, tom lane

--
Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

First  |  Prev  |  Next  |  Last
Pages: 1 2 3 4 5
Prev: Remaining Streaming Replication Open Items
Next: pgsql: Check compulsory parameters in recovery.conf in standby_mode, per