Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Say I have a bunch of clients with ONE master host that houses the gmetad and web interface. If I define one cluster (i.e. 'Cluster'), it works, but let's say I want a second set of cluster, how does that work? I can't point all my client nodes to the master server? It looks like I have to create some sort of "middle man" server to collect the data for each cluster?

share|improve this question
up vote 0 down vote accepted

My server with the web frontend and gmetad has gmetad.conf with one gridname, plus one data_source entry for each cluster. Each data_source draws from one or more systems in the cluster:

gridname "The Grid"
data_source "Infrastructure" ihost1 ihost2 ...
data_source "Compute Nodes" chost1 chost2 ...
data_source "Workstations" work1 work2 ...

Each host registers itself into a particular cluster, and onto a cluster-specific multicast address in its gmond.conf:

cluster {
  name = "Infrastructure"
udp_send_channel {
  mcast_join =
  port = 8649
udp_recv_channel {
  mcast_join =
  port = 8649
  bind =
tcp_accept_channel {
  port = 8649
share|improve this answer
oh, i have to use multi-cast? I've been using "host." Also, the address, that would be the gmetad host? – luckytaxi Apr 26 '11 at 17:56
I won't claim my setup is the only way it can work, but it's basically the default way that Ganglia is set up. Multicast may not be strictly required, but I suspect that the cluster {} group is. Maybe that's all you need to change. – Mike Renfro Apr 26 '11 at 18:16
I think my issue is I send EVERYTHING to the gmetad server. When I do this, everything goes into one cluster even though I define cluster {} on each node. – luckytaxi Apr 26 '11 at 18:31
Sounds reasonable. Again, this is how I've had it set up for years, and I never thought about doing it any other way (or if I did, I've long forgotten about it). – Mike Renfro Apr 26 '11 at 22:42

A slightly better variation on the configuration above would be to have multiple gmond's talking via multicast to a single gmond (for each cluster). Then setup the data_source to talk to that single gmond. That way, every time you add a new host, you don't have to edit the gmetad.conf.

share|improve this answer
Yep, doing this now and it works real well. – luckytaxi Jul 11 '11 at 11:59
if that single gmond goes down for whatever reason, you will lose statistics for the entire cluster. – user130073 Jul 27 '12 at 21:19

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.