the ganglia.info domain has been registered for the ganglia project.

The ganglia web page has been updated in order to make it easier to quickly find the information that you are looking for. The Ganglia Wiki has also been deployed.

I did a one-hour talk on ganglia at the Cluster World Expo 2004. About 80 people attended the talk and I got to meet some people face-to-face that have contributed to ganglia.

The Ganglia Development Team is please to announce the release of ganglia 2.5.6. To download this new release, please visit the ganglia download page. Changes (see ChangeLog for details): * The host dmax attribute is now honored allowing you to specify a timeout for dead nodes. See the host_dmax attribute in ./gmond/gmond.conf for details. [Federico Sacerdoti] * Ganglia will now compile on 64-bit versions of FreeBSD [Brooks Davis] * A bug has been fixed which caused nodes to falsely be marked as dead because of bogus TN attribute values. [Matt Massie] * An RPM dependency on librrd has been removed making installation and maintenance simpler. [Matt Massie] Thank you for using Ganglia! -The Ganglia Development Team

On December 17th, OffMyServer CTO Matt Olander and FreeBSD Project clustering expert Brooks Davis appeared on TechTV, a major technology network, for a segment entitled Building a Beowulf Cluster. The segment appeared on The Screen Savers show, which is one of TechTV’s most popular programs. TechTV is currently available in nearly 40 million homes in the United States and distributes content to more than 70 countries. Brooks Davis has made many contributions to ganglia and was kind enough to plug ganglia during the 6-minute segment

Ganglia is briefly discussed in Chapter 20 Instrumentation and Monitoring of The Grid2 book by Ian Foster and Carl Kesselman. There is also a nice screenshot of the web frontend.

Changes (see ChangeLog for details): * FreeBSD users will be happy to know that all standard metrics are now reported on FreeBSD thanks to the work of Brooks Davis. * This release also fixes a bug brought to our attention by Jim Prewett. This bug in gmond allows a specially crafted UDP packet to crash all gmond on a multicast channel with a segfault. This special UDP packet cannot be created using gmetric but requires a custom piece of code or a fuzzer. * The ganglia documentation is now included in the distribution as ganglia.html and README. We are working to make sure the documentation is simpler to use, up-to-date and directly answers the questions that you have about ganglia. Thank you for using Ganglia!

the ganglia development team is pleased to announce the release of ganglia 2.5.4. [Download Page] the largest improvements in this release are the result of the hard work of federico sacerdoti. he has modified gmetad to be interactive allowing for efficient primitive subtree queries of the xml tree. the web frontend has been updated to work with the interactive gmetad so you will notice very significant speed improvements when loading ganglia web pages. gmetad has also been improved to handle clusters with varying clocks (the graph timescales are no longer locked). the time and value thresholds for the gmond network monitors have been increased to prevent unnecessary network traffic. (ironic isn’t it?). martin knoblauch has also made ganglia more stable and correct on irix.

Ganglia made the front cover of the August 2003 edition of Linux Magazine. Forrest Hoffman did a writeup of ganglia monitoring in the Extreme Linux section.

this is primarily a bugfix release steve wagner submitted a patch that vastly improved the efficiency of gmetad. previously, a gettimeofday() call was made for every round-robin database updated whereas gmetad 2.5.3 doesn’t require any gettimeofday() calls. time is gathered from the XML. this may help users who experience gaps in their gmetad images. federico sacerdoti and tomas ogren (stric) tracked down a serious bug in gmond that could lead to stack corruption for long metric names (>32 characters). federico also made the cleanup thread more efficient and fixed a bug which prevented expired metrics from being deleted. martin knoblauch and jack perdue provided code to strengthen HPUX support. phil radden fixed a bug on linux which caused bytes_in and bytes_out to occasionally be incorrectly reported.