I just pulled the trigger on a new release which now means you can send collectl data directly to a ganglia gmond over a UDP socket. This has been testing on a 2000+ node cluster in which all the compute-node level gmonds have been replaced by collectl which is happily sending LOTS of data to a higher level gmond including Infiniband and Lustre data which are just 2 of the many types of data collectl knows how to collect. A side benefit of using collectl is you can have it log data locally on each node every 5 or 10 seconds while sending data to ganglia at a coarser rate, giving you access to more node level detail when you need it.
And since all instances of collectl synchronize to the usec the impact on fine-grained mpi jobs is reduced. If there is any interest, check out collectl at http://collectl.sourceforge.net/ and if you do have any questions feel free to raise them on collectl's mailing list or sourceforge forums. Of course you can always raise them on this mailing list too if you prefer. -mark ------------------------------------------------------------------------------ Crystal Reports - New Free Runtime and 30 Day Trial Check out the new simplified licensign option that enables unlimited royalty-free distribution of the report engine for externally facing server and web deployment. http://p.sf.net/sfu/businessobjects _______________________________________________ Ganglia-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ganglia-general
