[Beowulf] PXE/kickstart and 10GBase-T issues

[Joshua Baker-LePain]

Thanks to all who replied on and off list.  It looks like I took a big
step forward on this today.  We are in the process of transitioning
our cluster to jumbo frames, but not all parts of it have moved over
yet.  In the main section of the kickstart file for these nodes, I
used the "--mtu=9000" flag to the "network" directive, meaning that
the nodes used jumbo frames during the installation.  By removing that
flag and then modifying the interface config files in the ks post
section, I got all of the nodes to kickstart successfully using
standard size frames and then come up into production using jumbo
frames.  3 of the 48 nodes had SGE fail to start due to the network
not being ready, but that's something I can handle.

I'll have another batch of 48 nodes ready next week -- hopefully this
solution will work there as well.  If not, I'll report back that the
issue is still out there.

Thanks again for bearing with me.

-- 
Joshua Baker-LePain
QB3 Shared Cluster Sysadmin
UCSF