Thank you Steve, enabling igmpsnooping on our 3com switch allowed the nodes to talk over the default Multicast channel and the gmetad frontend now picks up the new nodes as gmond is started on them.
On another topic, slightly annoying, HyperThreading on the Intel machines falsely reports 4 CPUs when there are only 2 present. That skews the CPUs Total: number in Ganglia. Oh, well. Thanks, Ganglia is very useful and incredibly well designed and easy to use. Warwick Poole > > I'm no multicast whiz myself, but just off the top of my head: > > 1. Make sure IP multicasting is enabled in all your kernels. > 2. Make sure all your interfaces have multicast enabled (should show up > in 'ifconfig -a'). > 3. I have this same problem every time I bring a new subnet of hosts > online. I have to go my network team and tell them to enable "PIM" on > that switch. Protocol Independent Multicast. Once they do that, > everything just starts working. > > Steve Gilbert > Unix Systems Administrator > [EMAIL PROTECTED] > > > > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of > Warwick Poole > Sent: Thursday, November 18, 2004 12:22 PM > To: [email protected] > Subject: [Ganglia-general] Only single nodes working > > I am new to Ganglia (and IP Mulitcast) BUT I have searched around for an > answer and cant find one. > > Defintion of issue: Each host with gmond only sees itself and the > central host with gmetad does not receive or display information about > any other nodes, except for itself. > > I have around 100 Linux servers on the same subnet (10/100 ethernet). > They are not a GRID or cluster, simply very similar hosts performing > similar functions. There are multiple switches between hosts (different > models and manufacturers) but no routers. > > I have gmetad running on a central server (Fedora Core 2 with kernel > 2.6.5) with Apache/PHP/RRDTOOl setup and working fine. The web interface > shows the stats of the local nide perfectly. However, I have started up > gmond on other servers (mostly Fedora Core 1 with 2.6.1) but the node > information does not show up in the web interface on the central > server. > > I have checked the following: > > - gmond is running on all hosts and listening on 8649 > - I can telnet to 8649 on all hosts from gmetad server and receive XML > - gstat --all on each host only shows itself, even though all have the > same cluster name > > I suspect that since I know nothing about IP Multicast, that this is not > working. So I guess my question is, where do I begin looking? Should I > start with the kernels on the machines or the configs of the switches > (only some of them are Cisco and have any sort of management interface)? > Any info would be great. > > Thanks > Warwick Poole > > ------------------------------------------------------- > This SF.Net email is sponsored by: InterSystems CACHE > FREE OODBMS DOWNLOAD - A multidimensional database that combines > robust object and relational technologies, making it a perfect match > for Java, C++,COM, XML, ODBC and JDBC. www.intersystems.com/match8 > _______________________________________________ > Ganglia-general mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/ganglia-general
