Posts Tagged ‘cluster’

Oracle RAC Standard Edition to achieve low cost and high performance

Friday, November 28th, 2008

I finished today to create a new production environment based on 2 Linux serverX86_64 and running Oracle RAC 10gR2. (I know, there is 11g right now, but I’m a conservative!)
Wheeew, I just spent a couple of hours applying all the recommended patches!
We choosed 2 nodes with a maximum of 2 multi-core processors each one so we can license Standard Edition instead of Enterprise Edition. 64bits addressing allow us to allocate many gigabytes of SGA. I’m starting with 5Gb but I think we’ll need more. And a set of 6×300Gb 15krpms disks (it can be expanded with more disks and more shelves).
This configuration keeps low the total cost of ownership but achieves best performance.
Due to disks layout, costs and needed usable storage, we had to configure one huge RAID5 on the SAN with multi-path. I decided anyway to create 2 ASM disk groups (ASM is mandatory for Standard Edition RAC), one for the DB, the second one for the recovery area. With spare disks we should have enough availability and even if it’s a RAID5 I saw good write performances (>150M/s).

Welcome new RAC, I hope we’ll feel good together!

Tags: , , ,
Posted in Linux, Oracle Database, Performance | No Comments »

JBOSS Cluster isolation and multicasting

Monday, November 24th, 2008

I configured two JBoss clusters in the same LAN: a production and a test environment.
I decided to configure every single cluster with a dedicate private LANs using a restricted netmask to isolate production and test connectivity, so I assigned
192.168.100.0/255.255.255.0 to test and
192.168.200.0/255.255.255.0 to production.
I configured Apache and mod_jk to loadbalance activities between cluster instances.

The page UsingMod_jk1.2WithJBoss (http://www.jboss.org/community/docs/DOC-12525) is a good tutorial to achieve this.

What problems should I expect?
JBoss uses UDP multicasting to replicate informations across cluster nodes: even if I isolate TCP traffic, JBoss will “ear” messages sent from other clusters and will log a lot of warnings like the following:

… WARN [NAKACK] [...] discarded message from non-member ….

I had to change BOTH multicast ip address and port (attributes mcast_addr and mcast_port) in the following configuration files:

  • ./deploy/jboss-web-cluster.sar/META-INF/jboss-service.xml
  • ./deploy/jmx-console.war/WEB-INF/web.xml
  • ./deploy/cluster-service.xml
  • ./deploy/ejb3-clustered-sfsbcache-service.xml
  • ./deploy/ejb3-entity-cache-service.xml

Good luck!

Tags: , , ,
Posted in JBoss | 2 Comments »