On 10/5/05, Jeff Krimmel <[EMAIL PROTECTED]> wrote: > Greetings, > > I have an 18 node cluster (1 master node, 17 slave nodes), and I am > running the cluster on the 192.168.0 private network. The master node > is 192.168.0.1, and the slave nodes are 192.168.0.1xx. I have > installed ganglia-gmetad and ganglia-web on the master node, and > ganglia-gmond on each of the slave nodes. The only change from the > default installation of gmetad on the master node are the lines > > data_source "RayleighCluster" 192.168.0.1 > > gridname "RayleighGrid" > > trusted_hosts 127.0.0.1 > > all_trusted on > > The gmond is the default installation on the master node and all slave > nodes. All nodes have two network interfaces. The masternode's eth0 is > a connection to the outside world, and the slave nodes eth0 is used > only for booting via PXE off the master. All nodes' eth1 is used to > connect to the private network (192.168.0).
Does your config specify that eth1 is the ganglia mcast_if on all machines? What is the TTL on all machines, and does the problem go away if the TTL is increased? Did you remember to use gmond's port number on the source(s)? The Ganglia output can be seen here: http://rayleigh.cds.caltech.edu > It looks like all the data from all the nodes is lumped together. > Nothing happens when I select an individual node or try to change > metrics. How can I get Ganglia to display colored output of each node, > rather than showing the cumulative output 18 times. Try putting gmetad on all the hosts, or make sure you list each item seperately in gmetad.conf (see below). --- > 1. In the gmetad.conf try adding all hosts > Eg: data_source "RayleighCluster" 192.168.0.1 192.168.0.2 192.168.0.3 No, this means that the other hosts are used as a backup for the first one; it'd still get data from itself (192.168.0.1). How about: data_source "RayleighWeb" 192.168.0.1 data_source "Rayleigh1" 192.168.0.101 data_source "Rayleigh2" 192.168.0.102 data_source "Rayleigh3" 192.168.0.103 data_source "Rayleigh4" 192.168.0.104 data_source "Rayleigh5" 192.168.0.105 ... (etc)? BTW, IIRC, 3.0.x has a new config format, but the options and workings are the same otherwise, generally. (Location is now specified on the command line instead of gmond.conf; that took a while to figure out :/). No, it's not a known problamatic config; but I've never had 3.0.x on more than one host in the cluster (3.0.x on the web host, 2.5.x on the other machines -- don't ask why) myself. That said, 3.0.x. should work fine; I don't believe downgrading would help except for a change in the config layout. Your gmond configs look fine. I believe it's gmetad and the web interface which are at fault. -- ~Mike - Just my two cents - No man is an island, and no man is unable.
