Content-Type: text/plain;
  charset="iso-8859-1"
From: Richard Huxton <dev@archonet.com>
Organization: Archonet Ltd
To: Wei Weng <wweng@kencast.com>, pgsql-sql@postgresql.org
Subject: Re: concurrent connections is worse than serialization?
Date: Wed, 14 Aug 2002 10:18:38 +0100
User-Agent: KMail/1.4.1
References: <1029271170.16828.7.camel@Monet>
In-Reply-To: <1029271170.16828.7.camel@Monet>
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Message-Id: <200208141018.38875.dev@archonet.com>

On Tuesday 13 Aug 2002 9:39 pm, Wei Weng wrote:
> I have a testing program that uses 30 concurrent connections
> (max_connections =3D 32 in my postgresql.conf) and each does 100
> insertions to a simple table with index.
>
> It took me approximately 2 minutes to finish all of them.
>
> But under the same environment(after "delete From test_table, and vacuum
> analyze"), I then queue up all those 30 connections one after another
> one (serialize) and it took only 30 seconds to finish.
>
> Why is it that the performance of concurrent connections is worse than
> serializing them into one?

What was the limiting factor during the test? Was the CPU maxed, memory, di=
sk=20
I/O?

I take it the insert really *is* simple - no dependencies etc.

> I was testing them using our own (proprietary) scripting engine and the
> extension library that supports postgresql serializes the queries by
> simply locking when a query manipulates a PGconn object and unlocking
> when it is done. (And similiarly, it creates a PGconn object on the
> stack for each concurrent queries.)

I assume you've ruled the application end of things out.

- Richard Huxton