Howdy folks, I have a few Ganglia questions I hope someone can answer.
1. Has anyone had a problem with gmetad not failing over to an alternate data source? Ie: if I have a line like this in gmetad.conf: data_source "Linux Regress cluster" 172.16.208.247 172.16.208.250 ...if 172.16.208.247 crashes, all my graphs for this data source just seem to stop...it never fails over to the second host listed as it should. Anyone else have this problem? 2. I'm not clear as to what the following metrics mean: CPU_AIDLE PROC_RUN (total running processes?) PROC_TOTAL (total processes regarless of their state?) ...better yet, is there a place I can look myself for these definitions? There's nothing in the documentation, and gmond/metric.h and gmond/machines/linux.c don't mean much to me. Maybe the next release can have some explanatory comments in these files? 3. Is there a way with a single gmetad to group Ganglia into separate grids? I guess one ways is to simply set up another machine with another gmetad and just have two separate Ganglia environments, but I was wondering if it's possible to do this with a single gmetad? 4. Is it possible to configure Ganglia to not use multicast? Let me be more specific...if I have a subnet of 200 machines, I don't really need for each machine to know about the state of the entire subnet. What I would rather have is for all the machines to just talk directly either a couple of machines running gmond or gmetad. In my case (with everything on the same subnet), it just seems like a lot of network traffic for no reason. I can see how the multicast model can be extremely useful for many people, but I don't think it's needed in my environment. Can this be done with tweaking the "mute" and "deaf" settings in gmond.conf? ...I know this is a lot, but I hope somebody can take the time to help. Thanks!! Steve Gilbert Unix Systems Administrator [EMAIL PROTECTED]
