Ethernet Channel Bonding (ECB)...Revisited

[Timothy W. Moore]

My beowulf cluster is up and running (Thanks to all who helped!) and I 
am getting some nice performance with my software.  Of course, I am 
comparing the results to a Quad CPU shared memory server.  While running 
this weekend, node07 for whatever reason, stopped communicating and the 
run terminated in error.  I am using a single NIC in each node.

[1] If I had placed multiple NIC's in each node, would the communication 
route have been diverted and completed successfully?  In other 
words...In case of a flaky connection, does it provide an alternate 
means of communication?

[2] I am an engineer, not a computer scientist or IT guru.  I do not 
want to even think about re-compiling kernels for all those nodes.  So, 
I am running RedHat 7.2 out of the box.  If I add a NIC to each node and 
server, what is the method or command sequence that one must do to 
actually perform the bonding?  Or does the kernel see 2 cards and do it 
automatically? (I guess that would be a perfect world!)

Tim