Thanks Matthew. I favour your approach of grouping cluster data via gmetad. I'll give it a go ASAP. The deaf and mute modes now make more sense to me.
---- Yemi > -----Original Message----- > From: Matthew Massie [mailto:[EMAIL PROTECTED] > Sent: Wednesday, January 28, 2004 3:10 PM > To: Adesanya, Adeyemi > Cc: '[email protected]' > Subject: Re: [Ganglia-general] Is there no escaping multicasts??? > > > On Wed, 2004-01-28 at 11:51, Adesanya, Adeyemi wrote: > > Hi. > > > > It seems like I have tried just about every option in my gmond > > configuration but there appears to be no alternative to multicasts > > (sigh). I don't want every monitored node firing packets around the > > subnets. > > > > I thought that the list of trusted_hosts specified a group > of machines > > that would be unicast in addition to the multicast over the local > > subnet so I tried to be clever and set mcast_ttl to 0 so > the multicast > > would go no further than the host machine sending the message but > > alas, no data was sent to the trusted_hosts either. > > > > Any suggestions on where I can go from here? > > adesanya- > > it's easy to tell gmond to be "mute" or "deaf" in the > configuration file (/etc/gmond.conf). a "mute" gmond will > happily save all the data it receives on a multicast channel > but it will never send it's own data. a "deaf" gmond will > send all it's data via multicast but will never listen or > save data on the multicast channel. > > some people setup all gmond to be "deaf" accept one (or a > few) and then use that gmond to query info about the cluster. > > if you hate multicast altogether for some reason, you can get > around that too. > > set every gmond that you have to be both "deaf" and "mute". > then every gmond running on every host will only know about > itself. then use gmetad to pull all the data for your > cluster together (gmetad uses a TCP connection to the data > port of each gmond). you will need to manually specify every > host in your cluster in the /etc/gmetad.conf as a data source. > > hope this helps. > > good luck! > -matt > > p.s. the multicast traffic that ganglia generates is really > small and only hosts that join the multicast group even > process the packets (other hosts filter it out at the link > level). a cluster of 128 machines uses less then 56 kb/s .. > about the same as a dailup modem. most ganglia messages are > 8 bytes in size. > > > > > > > ---- > > Yemi > > > > > > > > > > ------------------------------------------------------- > > The SF.Net email is sponsored by EclipseCon 2004 > > Premiere Conference on Open Tools Development and > Integration See the > > breadth of Eclipse activity. February 3-5 in Anaheim, CA. > > http://www.eclipsecon.org/osdn > > _______________________________________________ > > Ganglia-general mailing list [email protected] > > https://lists.sourceforge.net/lists/listinfo/ganglia-general > -- > mobius strippers never show you their backside > pgp fingerprint A7C2 3C2F 8445 AD3C 135E F40B 242A 5984 ACBC 91D3 >
