15 messages in com.mysql.lists.clusterRe: Cluster performance with data nod...
FromSent OnAttachments
Giulio Ferro09 Aug 2007 13:05 
Giulio Ferro14 Aug 2007 05:19 
Jon Stephens14 Aug 2007 05:25 
Giulio Ferro14 Aug 2007 08:28 
Jon Stephens15 Aug 2007 00:27 
Giulio Ferro15 Aug 2007 11:02 
Stewart Smith16 Aug 2007 18:50 
Giulio Ferro21 Aug 2007 02:58 
Giulio Ferro25 Aug 2007 09:46 
Stewart Smith30 Aug 2007 20:53 
Giulio Ferro31 Aug 2007 07:57 
Stewart Smith02 Sep 2007 19:02 
Giulio Ferro04 Sep 2007 05:48 
Hartmut Holzgraefe04 Sep 2007 06:15 
Giulio Ferro05 Sep 2007 08:12 
Subject:Re: Cluster performance with data node down
From:Jon Stephens (jo@mysql.com)
Date:08/15/2007 12:27:10 AM
List:com.mysql.lists.cluster

Giulio Ferro wrote:

Jon Stephens wrote:

Giulio Ferro wrote:

No feedback anyone?

What are the specs on your cluster?

From my original message:

-----

Here's the setup: 2 sql node + 2 data node interconnected by a gigabit switch used just by the cluster. Mysql version : 5.1.20 Freebsd 6.2 stable amd64 (source update : 2 days ago) DataMemory : 6500M IndexMemory : 500M Tables are created in a tablespace

Under normal circumstances and after the first query, the performance is acceptable : standard select queries in my test db take about .3 / .4 sec to complete.

The problems start when I take a data node down (gracefully, with the command "<nodeid> stop"). When I do that the time to complete the same queries rises to an unacceptable 12 seconds and stays that way.

Then I start the node again and performances go back to what they were before.

Can any of the developers explain why this abnormal behavior occurs and what we can do to keep performance acceptable even in a degraded situation? (this is the whole point of a cluster, isn't it?)

Try repeating your test using 4 data nodes (and NoOfReplicas=2) to start with rather than just 2. I think you'll find when you drop one DN that your performance won't decrease nearly so much.

More data nodes -> greater parallelisation.

Conversely, 1 data node -> *no* parallelisation.

So what you're observing with 2 DN -> 1 DN isn't really "abnormal".

cheers

jon.

----

If someone in the list needs further information just ask...

--

Jon Stephens - jo@mysql.com Technical Writer - MySQL Documentation Team <!-- Coffs Harbour, NSW, Australia (GMT +10.00) --> <!-- Mobile: +61 (0) 402 635 784 --> Mobile (Sweden): +46 (0) 73 677 39 93 (GMT +02.00) MySQL AB: www.mysql.com