Re: PostgreSQL 9.0 x64 bit pgbench TPC very low question? - Mailing list pgsql-performance

From Jeff Janes
Subject Re: PostgreSQL 9.0 x64 bit pgbench TPC very low question?
Date
Msg-id AANLkTi=Mv2yNyKWvpdFwUTStH+znP1MNGhyLZ5byRqZH@mail.gmail.com
Whole thread Raw
In response to Re: PostgreSQL 9.0 x64 bit pgbench TPC very low question?  (tuanhoanganh <hatuan05@gmail.com>)
List pgsql-performance
On Sat, Dec 18, 2010 at 11:13 AM, tuanhoanganh <hatuan05@gmail.com> wrote:
> My app has ~ 20 exe file, each of exe create new connect to postgesql

But how often do they do that?  Does each exe make a new connection,
do one transaction, and then exit?  Or does each exe make one
connection, do one transaction, then close the connection and make a
new one?  Or does each exe make one connection, then stick around for
a while using that connection over and over again?

In the first two cases, indeed -C is the correct way to benchmark it,
but in the third case  it is not.

> and
> there are 10-30 user use my application so I need -C to check PostgreSQL
> performance.
>
> I will test without -C option. But is there any way to decrease connect time
> when there are 200 process, each of process will create new connect to
> postgresql.

I think the easiest way to decrease the connect time by a lot would be
use a connection pooler.

The critical question is how often does each process create a new
connection.  200 processes which make one connection each and keep
them open for 10 minutes is quite different from 200 processes which
make and break connections as fast as they can.


Cheers,

Jeff

pgsql-performance by date:

Previous
From: tuanhoanganh
Date:
Subject: Re: PostgreSQL 9.0 x64 bit pgbench TPC very low question?
Next
From: Ivan Voras
Date:
Subject: Re: PostgreSQL 9.0 x64 bit pgbench TPC very low question?