[HECnet] VCF-East on HECnet?

[Peter Coghlan]

Thankfully, we'd /probably/ be using something a wee bit faster than a 64K
leased line. ;)


As far as I recall, DEC originally required that a network interconnect cluster
required 10MB/s connections. I think they also said that all the nodes had to
be on the same ethernet segment with no repeaters or bridges intervening! Maybe
that was just a way of being able to blame the networking equipment if there
were clustering problems when the customers invariably ignored the rules...
I suspect usable clustering would probably need a couple of MB/s of stable, low
latency networking between the cluster nodes in both directions. In most
realistic cases, something would be needed at each end to encapsulate the
cluster traffic for transport over IP.

Clustering over slow or unreliable links is not pretty or fun. The results are
generally poor performance, with hangs, timeouts and machines falling over,
with all cluster nodes affected, not just the ones on the wrong side of the
link. VMS does clusterwide locking on almost all shared resources. If a node
can't reliably make contact with other nodes in the cluster, things start to
fall apart very quickly. Unlike DECnet, clustering is not designed to gracefully
cope with network problems.

Regards,
Peter Coghlan.