"Monitoring with Ganglia" Chapter 1 Introducing Ganglia

First, history
Ganglia from the University of California, Berkeley's "Millennium" project, has a history of 10几 years.


Second, Introduction
Ganglia is a cluster performance indicator (metrics) monitoring framework. The ganglia function is more powerful than similar frameworks, providing "quasi-real-time" performance data.
Ganglia is well suited for large cluster systems and is also applicable in small cluster systems.
Ganglia in order to fit into a large cluster system, the architecture of other monitoring frameworks is not the same.
The following scenarios apply to ganglia:
1) There is a computer cluster, you want to "quasi-real-time" monitoring the cluster, from which to obtain performance metrics data
2) You have no objection to maintaining a configuration file on each computer node of a computer cluster
3) The operating system and network in the computer cluster support UDP multicast and UDP point-to-point unicast
4) The cluster of computers consists of cluster collections, each containing one or more computers per cluster


Iii. composition
Ganglia consists of 3 types of service processes: Gmond,gmetad,gweb. The relationship between them is: independent of each other, each type of service process is not dependent on other classes of processes, mutual cooperation, in order to make the ganglia normal operation, the 3 types of service processes must be normal operation.
Figure 1 shows the relationship.

Figure 1

3.1, Gmond
Gmond installs and runs on each computer in a computer cluster, and it uses a configuration file that is stored on that computer, and the Gmond process interacts with the computer's operating system to obtain performance metrics data.
In one cluster, the Gmond process on each computer communicates with the Gmond processes on other computers, sends the performance metrics data it obtains and receives the performance metrics data obtained by other gmond processes, and the communication uses the "UDP multicast" mechanism. Since a cluster has only one multicast address, the final result is: In a cluster, each gmond process maintains a single copy of the performance metrics data for each computer in the cluster.
The Gmetad process wants to get the performance metrics data for all the computers in a cluster, and only needs to communicate with the Gmond process on a computer in that cluster.

The architecture of Gmond is shown in Figure 2 and Figure 1.

Figure 2

3.2, Gmetad
A Gmetad process to obtain performance metrics data for all computers within a cluster, simply configure at least one address information for the computer within that cluster in the configuration file used by the Gmetad process.
Allows multiple Gmetad processes to run, and each Gmetad process is responsible for monitoring the performance metrics data for a subset of the computers in the cluster.
The Gmetad process passes the obtained performance metrics data to RRDtool, which is saved by RRDtool to the hard disk.
The architecture of the Gmetad is visible in Figure 1.
3.3, Gweb
A gweb process can obtain performance metrics data through rrdtool and expose it through a Web UI. Gweb program is written in PHP language, so in order to run the Gweb process, the first to run a PHP-enabled Web server, such as Apache Web server.
The architecture of the gweb is visible in Figure 1.


Iv. Introduction to the contents of the chapters

The 2nd chapter introduces the installation and configuration of Gmond,gmetad and Gweb.

The 3rd chapter of

Describes extensibility. The 4th chapter introduces the function of Gweb in detail. The
5th chapter describes how to obtain extended performance metrics data. The
6th chapter introduces some solutions to the problem. The
7th chapter introduces the integration of ganglia with other monitoring frameworks. The
8th chapter introduces the support of ganglia to sflow technology. The
9th chapter introduces several real ganglia application cases.