new install - can't get past DHCP

[AFRL_floor]

We were given a LNXI cluster that had been 'nuked & paved' (std DOD
procedure). Thus we have lots of hardware with no OS or cluster SW.

We are attempting to boot compute nodes from a Centos 5.2 head node
and wireshark shows the request from the compute node and the offer
from the head node, but the compute node just keeps asking. Is there
special magic required in the DHCP server configuration to boot a
LinuxBIOS / etherboot compute node?

Other PXE-boot systems (desktops, laptops, nodes from other clusters)
seem happy with the DHCP offer and work fine. Anybody got any hints
for the guys who won the white elephant?

-Kevin Stroup

[Cameron Harr]

I don't know if it's possible, but you'd do well to get a hold of a copy
of Clusterworx, which will help you provision the nodes.

Assuming that isn't possible, one guess is that etherboot on node nics
is looking for a filename to boot from the dhcp server, although I don't
remember offhand the syntax. Perhaps someone with a current cluster can
help there.

If you want to do straight PXE, you might be able to go into the BIOS
and disable booting off the option rom and instead boot off the onboard
lan. This is assuming you even have the option of going into the main
BIOS (which depends on the type of node you're running).

Sorry I'm not of much help :(
Cameron

[AFRL_floor]

Thanks Cameron!

FYI - the 'magic' turns out to be the Vendor Encapsulated Option and
the REQUIRE_VCI_ETHERBOOT compile option to etherboot. Set it right
and the BIOS is happy. On the the ELFs...