IBM Power Support in Australia After End of Life

A Cluster Has to Tell Two Situations Apart

"My partner has failed" and "I cannot reach my partner" require opposite responses, and from a single node they look identical. Every clustering design is fundamentally an attempt to distinguish them, and every heartbeat path exists to give the node more evidence before it decides.

Guessing Wrong Is Worse Than Being Down

If a node concludes its partner has failed when the partner is alive and merely unreachable, both take ownership of the workload and both write to the same data independently. That produces corruption no failover mechanism is designed to survive, and it is caused by ambiguity rather than by any component failing.

Independence Is What Counts, Not Count

Two network paths through the same switch are one path with two names, and a cluster relying on them has the redundancy on paper only. Genuine independence — a network path plus a repository on shared storage, over different media — is what actually reduces ambiguity, and it is worth verifying by tracing rather than by reading the design.

A Failed Heartbeat Adapter Is Urgent While Nothing Is Wrong

The cluster keeps running perfectly and has lost the margin that keeps its decisions unambiguous. Nothing is operationally degraded, so the ticket ages, and the next event is judged on insufficient evidence. ICD supplies cluster adapters alongside POWER8 and POWER9 components past IBM end of support, tested and covered for one to three years.