[erlang-questions] Clustering through Multicast
Wed Apr 22 10:17:24 CEST 2009
May be the solution exists, but the idea just popped and want to clarify
if it really works this way.
Instead of the regular cluster node and unicast load balance to multiple
worker nodes, if we use the IP multicast to a group of servers with
replicated mnesia and let each node attempt to start a transaction and
only one node gets to act on it, based on the mnesia transaction write
lock, will there be less latency in distributing the transactions. And
also it might help in some cases wherein the delay in identifying dead
node would not effect the transactions within.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the erlang-questions