>> However, at times we would see heavy arp activity, which would choke all >> the 100MBPS. I'm not sure if it was mosix or something else, but since then, >> we changed lots of things, and can't be sure what was the cause.
> My network administrator noticed similar things coming from my small cluster. > After a bit of investigation, it seems that if a node goes down in a Mosix > cluster, all the other nodes arpflood with who-has requests. It became such a > pain that I only enable the cluster after working hours, so as not to flood > others' connections. Has anyone inquired about this on the mosix mailing list ? And if so, is there a way to fix it ? Cause, I would then like to run a mosix kernel all the time, instead of only after working hours. > While on the subject of mosix, same network admin mentioned that MPI/PVM may > run into trouble on the cluster. Any thoughts on that? MPI programs won't work, as mosix doesn't yet implement migratable sockets. I hear that the mosix guys are researching on this. PVM should work though, cause there are some MOSIX/PVM papers available from mosix.org. Of course, they couldn't have measured performance if it wasn't possible. ;-) viral -- Hanging on in quiet desperation is the English way, The time is gone, the song is over, thought I'd something more to say.
