Communications between node members:

Augment gethostbyname() to return IP addresses for nodes named in the
cluster namespace.

Add a library on top which provides optional reliable comms including
heartbeat.

Add a fault reporting layer which allows clients to request validation
of any peer-to-peer path through the fabric.

Question: does it really make sense to make "advanced" communications
primitives (such as recursive flood-fill broadcast+ack) available to
the integration layer?  These facilities are certainly useful to the
recovery mechanisms (especially for peer recovery in hierarchial
clusters), but their relevance to cluster integration is
questionable.