Deploying Oracle 9i RAC on AIX - Two Configuration Options
Abstract Oracle9i RAC requires special network configuration for cluster interconnect. In this hint and tip we discuss two configuration options you can use. For related information about this topic, refer to the following IBM Redbooks publication: Deploying Oracle9i RAC on eServer Cluster 1600 with GPFS, SG24-6954-00
Contents Option 1: Two private interconnects, with one primary and one standby
For this option, two private HACMP networks are used for the RAC interconnects. Both private interconnects should be defined under HACMP/ES with a "service" attribute. Oracle uses the HACMP/ES utility cllsif to determine the networks to be used for RAC interconnect. Since cllsif performs an alphanumeric sort of the network adapter names, Oracle will choose the first listed “service” adapter on the first “private” network for RAC interconnect. If the first private service interconnect fails, then Oracle Transparent Network Failover Failback (TNFF) fails over to the second private service interconnect. If both private interconnects fail, Oracle TNFF fails over to the public client network (the same alphanumeric sort order is used for “picking” the public network).
Option 2: EtherChannel with round robin over multiple interconnects
You can also use EtherChannel to bundle all the intended cluster interconnect adapters into one pseudo interface. The number of interconnect adapters required depends on the cache fusion requirements in your environment. The AIX implementation of EtherChannel supports up to 8 aggregate adapters per link. We tested the EtherChannel round robin algorithm. This can be used to evenly distribute the load over the links within the EtherChannel, which aggregates the bandwidth of all links. (The EtherChannel hash algorithm may skew the load to a certain link in a cluster if there are only a few nodes.) EtherChannel has high availability built in. Any EtherChannel link failure will be transparently handled by AIX. As long as one physical link is up, EtherChannel will continue to function. If the entire EtherChannel fails, Oracle TNFF will then fail the RAC communication to the public network.