I really don't understand this fear of multicast from managers. Unless you are running an old non-switched network or have igmp disabled, the small amount of multicast traffic that ganglia does produce will never even be seen by nodes that are not part of ganglia's multicast group. It also should never leave the subnet unless you have set a higher TTL and have your routers (or layer 3 switches) configured to pass multicast packets. Using multicast also eliminates a single point of failure for your monitoring system when you have a large cluster and it is normal for a few nodes to be down for various reasons. When that happens gmetad can just get the data from any of the other gmond since they all have redundant copies of the cluster data.
For our ganglia configuration, we use multicast without any problems and plan on upgrading to ganglia 3 soon and then will be using unicast basically only to allow nodes on different subnets to join an existing multicast cluster. We once played with the metric.h definitions a while ago to see if we could reduce the amount of multicast traffic, but overall significant changes could only be achieved by drastically changing the thresholds which is not desirable. Overall gmond's thresholds appear to be tuned fairly optimally. ~Jason On Tue, 2005-03-29 at 08:32 -0800, Jason Helfman wrote: > Hi. > > I have been looking through the mailing list for a little bit and > finally found some snippets about multicast. I am wondering if there are > any large issues in not using multicast, and going with unicast. > > My manager is not thrilled and not interested in using or supporting > multicast in ganglia, and I am wondering if this is going to cause an > issue in using the tool. > > I have only worked with it in a multicast environment, and never saw an > issue, however I don't know what issues there would be in a unicast > environment, if that is something that is supported. > > The snippets I found are: > https://sourceforge.net/mailarchive/message.php?msg_id=7047144 > > i"m sorry but ganglia 2.x relies on multicast for efficient group > messaging. we are definitely going to remove the reliance on multicast > (only) in the future but i know that doesn"t help you now. > > > https://sourceforge.net/mailarchive/message.php?msg_id=7047148 > The more I think about this, the more I"m convinced you are trying to > pound a square peg into a round hole. Multicast is an integral part of > ganglia. It really sounds to me like you need to run snmpd on each of > your machines and just query them from a central collector. You will > get the same information you would get with ganglia without the > multicast traffic. The big win with ganglia is being able to distribute > the data collection and you will lose that if you disable multicast. > > However, I"ve found that the multicast traffic I see here from > monitoring about 250 hosts is nearly unnoticable. I see more broadcast > traffic from the few Windows machines we have than from all of the > multicast combined. > > Thanks, > Jason > > > ------------------------------------------------------- > SF email is sponsored by - The IT Product Guide > Read honest & candid reviews on hundreds of IT Products from real users. > Discover which products truly live up to the hype. Start reading now. > http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click > _______________________________________________ > Ganglia-general mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/ganglia-general > -- /------------------------------------------------------------------\ | Jason A. Smith Email: [EMAIL PROTECTED] | | Atlas Computing Facility, Bldg. 510M Phone: (631)344-4226 | | Brookhaven National Lab, P.O. Box 5000 Fax: (631)344-7616 | | Upton, NY 11973-5000 | \------------------------------------------------------------------/
