Greetings. We are running a set of two-node Oracle RAC clusters for an upcoming ERP system - two 24x7 production clusters and three test clusters. Each RAC cluster shares its data/archive/quorum disks over FC using an EMC Symmetrix DMX, and each node of each cluster has its own dedicated Oracle Home filesystem. Due to the third-party software that is being managed on those Oracle Homes, the database folks want to be able to share Oracle Home between RAC nodes within a cluster. The application was not written with RAC in mind, and causes some file synchronization issues within the Oracle Homes that they want to eliminate.
I'm considering using one of the following Cluster Suite architectures. I have three test nodes available, each with a single-port FC card and connectivity to the EMC DMX. Each node would run 5.1 and GFS1 and implement "Managed NFS" as described by the Red Hat NFS Cookbook. I also have a WTI IPS800 network power switch for fencing.
In Option 1, one CS node would serve /orap1 and /orap2 to the two production clusters, The second CS node would serve /oratest1, /oratest2 and /oratest3 to the three test clusters. In Option 2, one CS node would serve /orap1, /orap2, /oratest1, /oratest2 and /oratest3 to the five RAC nodes on Oracle server "1" and the other node would serve those same five Oracle HOMEs to the five RAC nodes on Oracle server "2".
Given this, I have a few questions:
1. Does either testbed described above sound like the best architecture for this application or can somebody suggest a better way?
2. Would anyone share their experiences with sharing Oracle Home filesystems between RAC nodes using some form of Cluster Suite setup?
3. We are having internal discussions on the number of nodes that is adequate for this. Some in our area think a two-node active-passive NFS server running Heartbeat (without Cluster Suite) is sufficient. Since the NFS service will hopefully scale to allow us to replace other NFS servers, some think that Cluster Suite on three nodes will provide better overall availability and scalability. Given the fact that the Oracle RAC servers that will use the NFS service are 24x7 systems, some are concerned that even three nodes might be too few and that five nodes would provide better availability to handle scheduled maintenance, etc. Are there any guidelines anyone would like to share on this issue?
Thanks in advance,