Date: Fri, 23 Sep 2005 07:05:00 -0400
From: Michael Stone <mstone+postgres@mathom.us>
Subject: Re: How to determine cause of performance problem?
In-reply-to: <1127470875.3524.17.camel@Panoramix>
To: pgsql-performance@postgresql.org
Mail-Followup-To: pgsql-performance@postgresql.org
Message-id: <20050923110500.GX14918@mathom.us>
MIME-version: 1.0
Content-type: text/plain; charset=us-ascii; format=flowed
Content-disposition: inline
References: <1127458167.2475.105.camel@Panoramix>
	<20050923095517.GW14918@mathom.us> <1127470875.3524.17.camel@Panoramix>
User-Agent: Mutt/1.5.10i

On Fri, Sep 23, 2005 at 12:21:15PM +0200, Joost Kraaijeveld wrote:
>On Fri, 2005-09-23 at 05:55 -0400, Michael Stone wrote:
>> It's not clear what your object id generator does. If it's just a
>> sequence, it's not clear that you need this program at all--just use a
>> SELECT INTO and make the object id a SERIAL.
>It generates a GUID (and no, I do not want to turn this in a discussion
>about GUIDs). As in the Java code comment: it is not the generation of
>the GUID that is the problem (that is, I can generate millions of them
>per second.)

I didn't say it was, did I? If you use a SELECT INTO instead of
SELECTing each record and then reINSERTing it you avoid a round trip
latency for each row. There's a reason I said "if it's just a sequence".

>> If you do need to control the object id or do some other processing
>> before putting the data into the new table, rewrite to use a COPY
>> instead of an INSERT.
>It is actually the shortest piece of code that gives me a poor
>performance. The conversion problem is much, much larger and much much
>more complicated. 

Ok, that's great, but you didn't respond to the suggestion of using COPY
INTO instead of INSERT.

>But I have no clue where to begin with determining the bottleneck (it
>even may be a normal performance for all I know: I have no experience
>with converting such (large) database).
>
>Any suggestions?

Respond to the first suggestion?

Mike Stone