From: Ron Mayer <rm_pg@cheapcomplexdevices.com>
Subject: Re: Some performance numbers, with thoughts
Date: Wed, 28 Jun 2006 08:41:41 -0700
Organization: Hub.Org Networking Services
Lines: 27
Message-ID: <44A2A335.50601@cheapcomplexdevices.com>
References: <44973CC6.20309@janestcapital.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
To: Brian Hurt <bhurt@janestcapital.com>
User-Agent: Thunderbird 1.5.0.4 (Windows/20060516)
In-Reply-To: <44973CC6.20309@janestcapital.com>
To: pgsql-performance@postgresql.org

Combining the "insert" statements in a big concatenated string
joined by semicolons - rather than sending each individually
can drastically speed up your inserts; making them much closer
to the speed of copy.

For example, instead of sending them separately, it's much faster
to send a single string like this
  "insert into tbl (c1,c2) values (v1,v2);insert into tbl (c1,c2) values (v3,v4);..."
presumably due to the round-trip packets sending each insert takes.

Brian Hurt wrote:
> 
> Inserts, 1,000 per transaction             ~5,400 inserts/second
> Copy, 1,000 element blocks              ~20,000 inserts/second
> 

When I last measured it it was about a factor of 4 speedup
(3 seconds vs 0.7 seconds) by concatenating the inserts with
sample code shown her [1].

If the same ratio holds for your test case, these concatenated
inserts would be almost the exact same speed as a copy.

   Ron M

[1] http://archives.postgresql.org/pgsql-performance/2005-09/msg00327.php