Configuring a Cluster for Array Failover, Mirroring a Split Brain Detector Across Two Arrays in a Clustering Environment
Articles and Tips: tip
01 Feb 2001
If a node can't write to the Split Brain Detector (SBD) slot, it will take itself out of the cluster by assuming that whatever reason caused the I/O failure will also cause file systems and other applications on the node to also fail.
The idea is volumes and applications can then failover (quickly) to some other node that has a good I/O path. But, if you power down the array that contains the SBD LUN (Logical Unit Number), then all the nodes will detect an I/O failure and will also do this. To avoid this cascading effect, you have to mirror the SBD partition across two arrays (this is an option during install).
The same thing will happen if you power off the Fiber Channel hub, and you don't have two hubs. But, if you remove a Storage Area Network (SAN) cable and you only have one adapter, only that node will failover, so its volumes, etc., are then failed over to another (good) node.
It's a trade-off between cost and availability. You can always duplex everything: the HBAs, hubs, RAID controllers and even the array itself, if you want to eliminate every point of failure. For example, in some cases, the cost tradeoff wins because the arrays, have dual power; controllers, etc., and are already redundant.
Mirroring a Split Brain Detector Across Two Arrays in a Clustering Environment
To mirror a Split Brian Detector across two arrays in a clustering environment, perform the following steps:
Shutdown the cluster by typing CLUSTER DOWN.
Unload the cluster software on all nodes: to do this, type ULDNCS on all nodes.
Delete the existing SBD (Cluster) partition by using the NWCONFIG utility.
Create a new SBD partition by using the SBD INSTALL utility.
Type SBD INSTALL and follow the instructions. You will be prompted to chose the LUN upon which to create the SBD partition and whether to mirror that partition, and if you do decide to mirror, upon which to chose the second LUN.
You don't have to mirror the entire array, just mirror the 10MB SBD partition.
Remember, the SBD partition should be mirrored across a LUN in one array and the second LUN in the second array in order to protect the cluster against an entire array going down.
* Originally published in Novell AppNotes
The origin of this information may be internal or external to Novell. While Novell makes all reasonable efforts to verify this information, Novell does not make explicit or implied claims to its validity.