TRIUMF Grid Software and Applications

Ganglia Client Setup

All nodes need to run gmond.  This documentation has been updated for ganglia version >= 3.  To install gmond on a client do the following:

  • Install ganglia-gmond available at /opt/mirror/rpms/extras/ganglia:
    # cd /opt/mirror/rpms/extras/ganglia
    # rpm -iv ganglia-gmond-3*.i386.rpm
  • The rpm annoyingly immediately starts the service before you have a chance to configure it.  So stop it.  At the same time reset its runlevel so that it doesn't run at runlevel 2:
    # /etc/init.d/gmond stop
    # checkconfig --del gmond

    # checkconfig --level 345 on
  • Generate a configuration file.   It is possible to generate a version 3 file from an older version 2 gmond.conf file:
    # /usr/sbin/gmond --convert /etc/gmond.conf \ >/etc/gmond.conf.3
  • The following parameters need to be set in the configuration file /etc/gmond.conf:
    • cluster -> name - the name of the source group of elements that this node belongs to.  The current set of available names is:
      • LCG Servers -- misc. servers not members of the following:
      • LCG Storage Elements
      • LCG Compute Elements
      • LCG Workers
    • cluster -> owner - the top level name of the cluster (Triumf Grid)
    • udp_send_channel -> mcast_join - the multicast channel that this node should be a part of.  Do not use the default channel.  The channel addresses here are:
      • 239.2.11.70 -- LCG Servers
      • 239.2.11.72 -- LCG Workers
      • 239.2.11.73 -- LCG Storage Elements
      • 239.2.11.74 -- LCG Compute Elements
    • udp_send_channel -> mcast_if - the multicast ethernet interface for this machine.  It should be the interface on the private network if the machine is multi-homed.
    • udp_recv_channel -> mcast_join (like send channel above)
    • udp_recv_channel -> mcast_if (like send channel above)
    • acl -> trusted_hosts - the IP address of the ganglia server node.  Since lcfg is the server node you should use the private IP address:  192.168.1.250
      Obviously for a client node that is not on the private network you should use instead lcfg's public interface.
  • Once the configuration file is correct, launch the service:
    # /etc/init.d/gmond start