Hi all,
I'm running Ganglia on a grid with thousands of hosts, each with at least 4GB RAM installed. Since aggregating the total amount of RAM exceeds 4294967296 (the max value of uint32), I'm getting incorrect data on the total memory in the grid. I've peeked at gmond and gmetad code and adding a uint64 doesn't seem trivial. I've also searched the Net for anyone that's possibly encountered this before and came up only with this: http://sourceforge.net/mailarchive/forum.php?forum_id=9584&max_rows=25&style=nested&viewmonth=200312 It mentions that Ganglia 3.x should solve the problem. I'm running 3.0.2, but still experiencing it. I've also failed to find any method for changing collection of memory metrics from KB to MB, other than modifying the source code, which I'd rather avoid so as to keep as close to the original tree as possible. My questions are: 1. Is there really a solution in Ganglia 3.x and if so, what is it? 2. If not, are you aware of anyone who's implemented such a solution or documented the work needed, so I may carry on from there? Note that the gmetad collectors are running on a SuSE x86_64 machine and were compiled with 64bit libs. Ganglia is deployed in a hierarchy and reporting is done via unicast. In essence, this isolates the problem to the gmetads only, as the gmonds report on groups with less than 4TB RAM (but it may definitely surface there someday as well). Thanks, Alex
