>>> On 5/27/2008 at 8:47 PM, in message <[EMAIL PROTECTED]>, randy <[EMAIL PROTECTED]> wrote: > Brad Nicholes wrote: > >> Is there a reason why you would want to list all 120 nodes in the >> data_source directive of gmetad? When you list multiple modes in a >> data_source directive, it does not mean that gmetad is pinging all of >> them for data. It simply means that if gmetad can not get data from >> the first one in the list, it tries the next one. > > Only because I thought I needed too. I'm doing the multicast, but I just > thought that all the nodes had to be listed. > >> architecture, data is pushed from a monitored node to either a >> multicast channel or to a single gmond master node. A gmetad >> data_source simply references one node that is listening on a >> multicast channel or the single gmond master node in the case of >> unicast. There should never be a need to have more than just a few >> nodes listed in a data_source. > > So if all 120 nodes are talking on the multicast address, and gmond is > running on a different node on the same net, I can get by with just > giving localhosts as the data_source? Or any one (or more) of the data > nodes? I think that's how I left it today, and I was seeing reports from > 30 of the nodes and the other 90 were listed as down. The gmond machine > is also serving the webpages, and all machines can see each other on the > network. > > Appreciate the help, thanks Brad! > > randy
I would suggest that you break up your 120 nodes into separate cluster on different multicast channels. Then you would have a different data source for each cluster. By putting them all on the same channel, every gmond agent is required to store all of the metric data for all 120 nodes. That is a huge waste of memory. I would suggest breaking them up in to smaller clusters. If that doesn't work for you, then you might want to move to unicast rather than multicast. In unicast mode all of your gmond agents talk to one or more gmond master nodes directly rather than on a channel. Then each of the master gmond nodes becomes a separate data source. Search back through the email list archive for similar questions. Optimal Ganglia architectures have been discussed previously including multicast vs unicast. Brad ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Ganglia-general mailing list Ganglia-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-general
