Hello Peter,
On 2/18/2011 12:38 PM, Peter Sørensen wrote:
> Hi,
>
> I have a cluster with 2 API/SQL nodes and 1 managernode. Version used is 5.1 (7.09
> Ubuntu).
> The data nodes have 16G Memory
You don't say how much data nodes, and how much physical machines you
run this setup on. From what I can tell: 2 machines for data nodes, 2
for API/SQL/MGM nodes, so at least 4 machines?
Also, MySQL 5.1 should not be use for MySQL Cluster. You should use
MySQL Cluster 7.1, and our binaries/source.
> Among other things I use this as a service backend for my mailscanner gateways.
> They use spamassassin MySQL based. This I have implemented as a NDBCLUSTER DB but
> I have terrible performance problems using this.
>
> The estimates size of the DB is around 4G. The table I query has around 25 Million
> records.
> So should I go back and use InnoDb for this as I did in my prior setup or do anybody
> have some tuning recommendations.
Yes, it might be good to go back to InnoDB if the application isn't
written for MySQL Cluster. Maybe you can check if every query is using
an index for starters, and limit the use of joins (although with 1 table
that might hard :-)).
I'm not sure you would need HA like MySQL Cluster provides for
spamassassin data, which might just be good enough using MySQL
Replication. Just an idea.
Cheers,
Geert
--
Geert Vanderkelen
Oracle, MySQL Principal Support Engineer
ORACLE Deutschland B.V.& Co. KG
Registered Office: Riesstr. 25, 80992 Muenchen, Germany
Commercial Register: Local Court Of Munich, HRA 95603
General Partner: ORACLE Deutschland Verwaltung B.V.
Rijnzathe 6, 3454PV De Meern, Niederlande
Register Of Chamber Of Commerce: Midden-Niederlande, No. 30143697
Managing Directors: Juergen Kunz, Marcel van de Molen, Alexander van der Ven