Ganglia is a real time cluster monitoring tool that uses a protocol based on XML and XDR across a multicast network to provide a scalable core tool set. The cluster monitoring data can be accessed via a web browser or via the command line. The core Ganglia monitoring toolkit also includes the ability to add user defined data to the XML streams.
Ganglia was developed at the University of California, Berkeley
Computer Science Division as part of the ongoing clustering research
project named The Millennium Project (and its predecessor The NOW
Project). It is being developed by Matt Massie
mailto:massie@cs.berkeley.edumassie@cs.berkeley.edu
, Brent Chun
mailto:bnc@caltech.edubnc@caltech.edu
, Steven Wagner
mailto:swagner@ilm.comswagner@ilm.com
,
Federico Sacerdoti
mailto:fds@sdsc.edufds@sdsc.edu
, and other active developers.
For additional information about the Ganglia project, see:
ganglia-webfrontend is a package in OSCAR that sets up the web page display on the head node of the data gathered by the gmond and gmetad tools in the ganglia-monitor-core. This uses rrdtool package included with the OSCAR tool suite as well as the apache, php, and expat packages installed with the core Linux distribution on the head node to build an interactive useful display of the XML/XDR data being passed across the multicast channel.
Once the cluster head node has the OSCAR tools installed a web browser can be pointed at:
to see the ganglia web output. If nothing appears, check to see if apache has been started along with the gmond and gmetad processes. The start-up init scripts can be found in /etc/init.d/.
The 2.0 version of OSCAR also includes the ganglia python command line interface so access to the ganglia XML/XDR data can be scripted. This is a tool contributed to the Ganglia project by the Rocks team from the San Diego Supercomputer Center. Since it is based on a python class file this will depend on having python and the PyXML-oscar RPMs installed on the head node.
The current version of the Ganglia tools packaged for OSCAR is:
The addition of the ganglia command line tool makes it possible to access all of the available multicasted ganglia monitoring data that is flowing across the cluster from the command line or various perl, shell, or other system admin created scripts. The tool can be found in:
and the actual python class file gets installed in:
Running this tool on the command line gives the following usage message:
usage: ganglia < metric > [ metric metric ... ]
where ``metric'' is one of:
So for instance, the user can do:
# ganglia swap_total cpu_speed cpu_num
to produce the following output:
startx 265064 851 1 xcvs 1582384 733 2
Also included in the extras directory is a small example script to add
some network statistics to the ganglia data stream.
This script was developed by Goneri Le Bouderxi
mailto:glebouder@mandrakesoft.comglebouder@mandrakesoft.com
of MandrakeSoft.
Goneri's ganglia-network.sh script can be run periodically by the cron system to add network statistics to the Ganglia data stream. This is just an example script that is not officially part of the OSCAR cluster tool suite. No guarantees are included that this script will work in current or future versions of Ganglia that may be packaged as part of the OSCAR cluster tool suite.
root 2002-11-08