Re: Can anyone explain this pgbench results? - Mailing list pgsql-performance

From Joost Kraaijeveld
Subject Re: Can anyone explain this pgbench results?
Date
Msg-id A3D1526C98B7C1409A687E0943EAC410605FC3@obelix.askesis.nl
Whole thread Raw
In response to Can anyone explain this pgbench results?  ("Joost Kraaijeveld" <J.Kraaijeveld@Askesis.nl>)
Responses Re: Can anyone explain this pgbench results?
List pgsql-performance
Hi Michael,

Thanls for your response.

Michael Fuhr wrote:
> On Mon, Mar 06, 2006 at 04:29:49PM +0100, Joost Kraaijeveld wrote:
>> Below are some results of running pgbench, run on a machine that
>> is doing nothing else than running PostgreSQL woth pgbench. The
>> strange thing is that the results are *constantly alternating* hight
>> (750-850 transactions)and low (50-80 transactions), no matter how
>> many test I run. If I wait a long time (> 5 minutes) after running
>> the test, I always get a hight score, followed by a low one, followed
>> by a high one, low one etc.
>
> The default checkpoint_timeout is 300 seconds (5 minutes).  Is it
> coincidence that the "long time" between fast results is about the
> same?
I have not measured the "long wait time". But I can run multiple test in 3 minutes: the fast test lasts 3 sec, the long
one40 secs (see below). During the tests there is not much activity on the partition where the logfiles are (other
controllerand disk than the database and swap) 

postgres@panoramix:/usr/lib/postgresql/8.1/bin$ time ./pgbench  -c 10 -t 150 test
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 100
number of clients: 10
number of transactions per client: 150
number of transactions actually processed: 1500/1500
tps = 531.067258 (including connections establishing)
tps = 541.694790 (excluding connections establishing)

real    0m2.892s
user    0m0.105s
sys     0m0.145s


postgres@panoramix:/usr/lib/postgresql/8.1/bin$ time ./pgbench  -c 10 -t 150 test
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 100
number of clients: 10
number of transactions per client: 150
number of transactions actually processed: 1500/1500
tps = 37.064000 (including connections establishing)
tps = 37.114023 (excluding connections establishing)

real    0m40.531s
user    0m0.088s
sys     0m0.132s

>What's your setting?
Default.

> Are your test results more consistent
> if you execute CHECKPOINT between them?
Could you tell me how I could do that?



Groeten,

Joost Kraaijeveld
Askesis B.V.
Molukkenstraat 14
6524NB Nijmegen
tel: 024-3888063 / 06-51855277
fax: 024-3608416
e-mail: J.Kraaijeveld@Askesis.nl
web: www.askesis.nl

pgsql-performance by date:

Previous
From: Michael Fuhr
Date:
Subject: Re: Can anyone explain this pgbench results?
Next
From: "Matthew Nuzum"
Date:
Subject: Re: Postgres on VPS - how much is enough?