Cluster VS. HACMP or HADR?

Cluster is a general term about putting together a bunch of computers closer than a normal network does.

It involves 2 major purposes

• For high availability
• For more computing power

or both.

HACMP (High Availability Cluster Multi-Processing) is IBM AIX's implementation of cluster. It's an OS level high availability solution. When H/W or OS fails, the standby system would take over network connections and resume host identity as the original one.

But after and on top of that, you still need subsystems/middleware/service to work.

Some simple services are easier to resume than others. Those without 'memory effect' are the easier ones, such as HTTP server. The standby HTTP server with access to the same file system would start serving subsequent requests as soon as DNS updated or IP address assumed by the new master AIX.

For complex services such as DBMS, you'll need to deal with those on-the-fly transactions. For simple clustering, the standby (monitoring) DBMS would have to perform backout, lock-release, and recovery tasks before takeover could proceed. For higher level clustering, data sharing and hence workload sharing is implemented, and one DBMS down only means less processing capacity.

Some vendors like SteelEye combine OS and subsystem features together to provide a more coordinated high availability clustering.

IBM DB2 HADR (High Availability Disaster Recovery) is an alternative. It provides online replication function to prepare a full copy of db data somewhere else. When the 2nd copy is close to the primary one, it's a high availability solution; if it's far far away, then it becomes a disaster recovery solution.