Richard,
You are mostly correct. If you unicast to two nodes, you are saving on the UDP
traffic that you are sending. But you do save on the UDP traffic you are
receiving. If "X" is the amount of traffic generated by a single node sending
to one host, and you have "N" nodes in the cluster, then:
Unicast (to two hosts):
Send = 2 X
Receive = 0
Multicast:
Send = X
Receive = N X
Overall, the total amount of traffic (send + receive) is less for the unicast
case. What you need to decide is just how big N x X will be, and if that amount
of traffic will adversely affect anything. In most cases, there aren't any
problems, but it's one of those things that depends heavily on each individual
setup. So the "correct" choice is whatever fits your personal needs for
redundancy and network traffic load.
The unicast approach does save on gmond memory usage as you mentioned. It's up
to each site to determine just how much memory the metrics will take up, and if
it is considered a significant amount. (But it can get somewhat big on a
large cluster like mine with a bunch of added metrics.)
-- Rick
--------------------------
Rick Mohr
Systems Developer
Ohio Supercomputer Center
On Mon, 30 Jan 2006 [EMAIL PROTECTED] wrote:
Multicast.
If you unicast to 2 separate nodes for resiliency, then
you are actually sending double the UDP traffic to a multicast solution.
So unicast does not save on UDP network traffic - the opposite actually
in the resiliency
case. What unicast does do of course is reduce the load and memory
footprint
of most of the individual gmond processes, as they are not taking care
of the
state of the whole cluster.
Dear Gurus, I hope this paper analysis is correct.
regards,
richard
