Introduction to Ganglia

The Ganglia project was started at the University of California, Berkeley to support scientific research that depends on computers for data analysis. The Ganglia web page says the project "strives to advance science by creating a ubiquitous, continuous, and pervasive national computational infrastructure: the Grid." For the moment, however, we are only concerned with Ganglia's ability to monitor and display information about the cluster we have built.

Note 

Like the term cluster, the term grid has been used to describe a variety of computational systems. The most common use of the term evolved from the scientific community where it was used to describe multiple independently managed geographically disparate computing clusters. Using this definition, a grid can contain a cluster, but a cluster does not contain a grid. Both a grid and cluster, however, are computational environments where workloads run in parallel.

The Ganglia package consists of several command-line utilities, which we'll see shortly, and daemons that run on the cluster nodes. Let's briefly examine two of the Ganglia daemons before I describe how to install Ganglia on the cluster.

gmond

gmond is the Ganglia monitoring daemon. gmond's job is to gather performance metrics on the machine it is running on and to keep track of the status of every other gmond daemon running in the cluster. If one of the gmond daemons dies (due to a failure of the cluster node, for example), all surviving gmond daemons find out about it.

The gstat utility can report the information collected by gmond in XML format. Starting with version 2.7.0, gmond can compress this XML data before it is sent over the network.

gmetad

The gmond daemon is required on every cluster node, but the gmetad daemon only needs to run on the cluster node manager.^[1] The gmetad daemon polls the gmond daemons for their performance metric information every 15 seconds and then stores this information in a round-robin database using the RRDtool. (In a round-robin database, the newest data overwrites the oldest data, so the database never fills up.)

gmetad displays the information it collects using the Apache web server. Its ability to display web pages is made possible by the Ganglia Web package (formerly called the Ganglia Web Frontend). We'll discuss Ganglia Web shortly.

^[1]See the introduction for a description of the cluster node manager.