Re: Experiences with clusters on frequently joined data

From: Guy Harrison <gharriso_at_werple.net.au>
Date: 1997/07/14
Message-ID: <01bc9011$a53096c0$0100007f@gharriso>#1/1

Eric,

The small amount of research I've done on clusters suggests that while they markedly improve join performance, the impact on scans of individual tables can be severe.

If you do create a cluster, think very carefully about your setting for the SIZE parameter. This controls the amount of space reserved for each cluster key. If set too high, the cluster will be sparsely populated and table scan performance will suffer. If set too low then blocks with the same cluster key will chain and this will degrade lookups on the cluster key.

If the size of your cluster changes you should adjust the setting for SIZE and rebuild.

-- 
Guy Harrison

gharriso@werple.net.au || http://werple.net.au/~gharriso || 613 419377964



Eric A. Stephens <estephen_at_javaboy.appliedtheory.com> wrote in article
<m3en982rg1.fsf_at_javaboy.appliedtheory.com>...


> Would anyone care to comment on the pros/cons of using clusters

> to gain performance on frequently joined data. We certainly

> want to avoid excessive denormalization in order to gain 

> performance and we view clustering as a viable alternative

> -- 

> Eric A. Stephens                      estephen_at_AppliedTheory.com

> Software Engineering Group          http://www.AppliedTheory.com

> AppliedTheory Communications, Inc.

>

Received on Mon Jul 14 1997 - 00:00:00 CDT