Cerebro is a collection of cluster monitoring tools and libraries.
This project has been named after a part of the central nervous system
to pay homage to the the Ganglia project
(http://ganglia.sourceforge.net).  

A heavily modified version of Ganglia was once used at Lawrence
Livermore National Laboratory, but as needs and demands changed, it
became necessary to develop a slightly different tool.

Several of goals of Cerebro were to develop a monitoring tool that:

- Interrupt the CPU less frequently (fixed in newer Ganglias).

- Provide the user with a set of libraries, command-line tools, and a
dynamic module interface that allow users to monitor new metrics
without any re-compilation or configuration changes.

- Provide a dynamic module interface that allows individual clusters to
monitor different metrics based solely on the modules installed.

- Provide a dynamic module interface which allows the libraries and
tools to have knowledge of every node in the cluster.

- Provide a dynamic module interface which allows the libraries and
tools to automatically configure itself across a cluster.

- Define what nodes exist within a cluster.

- Remove XML overhead.

At this point in time, the Cerebro project should not be mistaken for
a full host monitoring system like Ganglia or Supermon, although it
one day may be.  

Future development goals include:

- Security/access control features

- Confined listening configuration.

In order to install Cerebro, only the primary RPM (cerebro) need be
installed.  However, additional features of Cerebro can be taken
advantage of by installing a clusterlist module (cerebro-clusterlist)
which allows Cerebro to recognize the hosts in the current cluster.

The current clusterlist database formats supported are:

Genders - Uses a genders database and the genders library to determine
all nodes in the cluster. (http://sourceforge.net/projects/genders)

Hostsfile - Uses a file that lists the hostname of every node in the
cluster on a separate line.

In addition, other modules can be installed for cluster configuration
(cerebro-config), metric monitoring (cerebro-metric), or metric
logging (cerebro-monitor).

The following metric modules are currently publicly distributed:

boottime - Monitor cluster node boottimes.

shutdown - Monitor when nodes are shutdown gracefully

bytesin, bytesout, packetsin, packetsout, rxerrs, txerrs - Monitor network usage

memtotal, memused, memfree, swaptotal, swapused, swapfree - Monitor memory usage

loadavg1, loadavg5, loadavg15 - Monitor load

Albert Chu
chu11@llnl.gov
Lawrence Livermore National Laboratories