I have fiddled to my hearts content and can't get them seeing other nodes.
That being said we have investigated very thoroughly this issue. The
only thing we can come up with at this point is the Opteron RHEL WS 3
seems to be using IGMPv3, while all of our other machines are using v2.
There are some include/exclude options with IGMPv3 that allows
filtering.. so we are looking more into that.
Does anyone know if there is a way to force down to IGMPv2 either at the
machine level or application level?
Thanks again,
Pete.
Joshua Baker-LePain wrote:
On Thu, 10 Jun 2004 at 8:25am, Peter Schmid wrote
The Xeon nodes all see one another.. and work fine. The opteron nodes
only see themselves. Using tcpdump, it sees no other mcast traffic on
the mcast IP other than itself. It's not a issue on the switch... and I
can telnet to the mcast port on each machine and see only that machines
data... so it's not a blocked port. And again... each cluster is
attached to the same switch.
Is anyone else using this with RH AW 3?.. and/or opterons. I'm stumpped.
Have you tried bouncing ganglia on the Opterons? I *am* running ganglia
on Opterons running RHEL WS 3, and I vaguely recall having this issue.
IIRC, I just fiddled with stuff, mainly bouncing ganglia I believe, and
got it to work.
--
Peter Schmid
Technical Director
Logic Technology Inc.
GE Global Research Center
ITMS Engineering Systems Group
(518) 387-6903
