On Wed, Apr 15, 2009 at 04:51:57PM -0400, Perry E. Metzger wrote: > > In our brand new cluster, we're using Intel S3210SH motherboards. > > The boards are going to be managed by a pure hands off system I've > built. IPMI is used for tasks like monitoring and telling the boards to > PXE boot so they can be re-installed by a purely automated system when > software upgrades happen. > > Unfortunately, every once in a while, the IPMI BMCs on my test systems > simply stop talking to the network. This isn't overly tragic since I can > have a process go over to such a board when it detects that pings have > stopped working and use a local IPMI command to cold rest the BMC, but > it is still really Not The Right Thing. Also, I suspect every once in a > great while I'll get a simultaneous OS and IPMI BMC failure and shoe > leather will be needed to reset the box, which I don't like.
We also had this problem with Supermicro boards and IPMI cards in a large scale. Finally we found a solution by upgrading the firmware of the NICs which are actually from Intel. You might want to ask the vendor or you trader to get a beta version which is more recent than the public available one for both - the IPMI cards and the NICs. Unfortunately, this problem occurs occasionally which makes testing difficult. We took a supset of nodes, played the new NIC firmware onto it and waited a long time. Good luck, Henning _______________________________________________ Beowulf mailing list, [email protected] sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
