[Linux-cluster] RHEL5.3 Cluster - backup fencing methods

[Kaerka Phillips]

Hi - I've got an issue with a 4-node cluster, and I'm hoping to get some
good advice or best-practices for this.  The 4-node cluster is on dell
hardware, using DRAC cards as the primary fencing device, but I'd like to
eliminate the single-point of failure introduced with the cabling for this
method.

I attempted to use Fence_ipmilan, but once i got the fence_drac5 working,
this no longer works for unknown reasons, but even if it did work, the DRACs
are on a private VLAN, as is the cluster and cluster multicast address.  I'm
concerned that a failure of the switch which hosts that vlan and drac
ethernet connections would cause an outright cluster failure.  The point of
this cluster is to share GFS2 filesystems amongst the 4-nodes.

My network setup is this:
2x GigE cables plugged into PCI-E cards (two physical cards), bonded
ethernet config, public network
1x GigE cable plugged into port #1 on dell server, in "shared" mode in DRAC5
card, on private network:
     DRAC5 card maps to this with an assigned ip address, OS also maps to
this with a different assigned ip address.

All cluster communication between nodes passes over private network over
GigE shared port.  I've not been able to determine the correct solution to
eliminate this last single-point of failure, aside from adding an additional
network connection to another on-board ethernet card, and mapping this to
the private network, to a 2nd switch.  I have no guarantee that this will
work, nor much documentation to indicate what setup would be required for
this.

Any thoughts?

Thanks,

Kaerka
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/linux-cluster/attachments/20090402/d8e87000/attachment.htm>