|
Has anyone looked into duplicating the cluster management/HA features
of TurboCluster? I don't really know the history of the divergence of
TurboCluster from the lvs project, so I hope I don't start any
flamewar by mentioning their name.
To me, their approach is very appealing. It looks like each node in
the cluster runs a cluster management daemon, which allows you to
easily synchronise the cluster configuration, and it also provides HA
by electing a new master Director if the master dies. I assume they
use the "Bully Algorithm" to do the election, as it is a well known
algorithm for doing an election in a distributed environment..
I think this really simplifies cluster management and HA. I'd much
rather use their approach than have to fiddle with separate packages
like mon and heartbeat.
One advantage to their solution is that all the real servers can also
be directors or backup directors. For a small outfit, this would seem
to maximize ones hardware value.
I think this would be a fun project, and I might be able to spare a
couple of weeks of time to start a cluster manager prototype in perl.
Cheers,
Doug
|
