Oracle RAC Standard Edition to achieve low cost and high performance

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 6x300Gb 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!

JBOSS Cluster isolation and multicasting

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!

It’s time to trouble…

Sometimes it’s hard to find enough time to write something or even to only THINK about writing something

The following are the projects I have to complete before the deadline of December 17th (at least if I still want to go on vacation…)

  • A totally new Oracle 10gR2 RAC SE on Linux (OCFS2, ASM) including jboss frontends, backups, monitoring, documentation. (Servers are ready today).
  • A Disaster recovery architecture based on Dataguard with scripts based on rsync to do filesystem replication, with failover and failback, including backups, monitoring, documentation. (The server in DR site is reachable via network today).
  • A 17 server infrastructure (among others a RAC 10gR2 on linux) transfer from Milan datacenter to here. It’s planned for december 11th but I have to crosscheck backup and contingency requirements.
  • A 14 server infrastructure (based on Windows and SqlServer) transfer from Milan datacenter to here. To be planned in december.
  • A totally new cold failover cluster based on linux with Oracle DBMS and E-business suite (Servers will be provided soon, I hope!).
  • A new standalone Windows Server 64bit to outstand the 32bit allocation bottleneck for a 500Gb oracle database (Server will be provided not before december 10th).
  • Normally manage the day-by-day work, including replying to e-mails and answering the phone.

AARGH!!