OpenSSI cannot find NIC with static node configuration

scrottie on 2010-07-14T18:10:34

Pasting an email reply since everyone seems to be stumbling over this one:

----

I've had this problem too.

This happens when the driver for the NIC on that machine isn't included in /etc/mkinitrd/modules. When the node boots, it can't find the NIC card because it doesn't have a driver for it, and since it can't find the NIC card, it doesn't know the MAC address, and without the MAC address, it can't configure the node. Without the NIC card, it can't connect to the initnode.

You can add the names of the NIC devices for all of the machines during install, or you can add them before you boot the other machines into the cluster and then run 'mkinitrd -o /boot/initrd.img-2.6.14-ssi-686-smp 2.6.14-ssi-686-smp' (adjusted to match your kernel version, see 'uname -a') and run 'ssi-ksync'.

It's possible to add the correct device names and rebuild initrd but still the node won't find the NIC or MAC address. For some reason, some drivers just don't work correctly with how OpenSSI configures nodes. The e100 and e1000 don't work correctly with OpenSSI for configuring nodes. I got a bunch of 3com 3c509 cards and stuck those in the nodes and use only those for the cluster interconnect, for now. Later I hope to get Infiband going.

So, make sure /etc/mkinitrd/modules has the correct driver in it, rebuild initrd, ssi-sync, and if all else fails, use some old 3com cards.

Good luck! -scott