RAID in clusters

The practice of clustering systems has increased in popularity over the past few years, and appears to be gaining momentum. This is not surprising, considering that clustering offers an alternative to very large computer systems that is both cheaper and more reliable.

Reliability is one of the major benefits of building clusters. If all the systems in a cluster need to be online for the cluster to operate, the Mean Time Between Failures (MTBF) would be unbearably short and reliability poor. This is analogous to using RAID 0 in a data storage array. The MTBF of a RAID 0 group with 10 drives is 10 times less than that of a single drive-because if one drive fails, the entire group goes offline. The same would be true for clusters without redundancy.

Cluster Redundancy

The many different clustering applications solve the reliability problem in the same way-software that allows one of the other Nodes in the cluster to take the place of a failed Node. This is only possible if both Nodes can access each other's disk drives. As an example, consider a 2Node cluster running two databases-one on each Node. If Node A goes down, Node B needs to be able to access Node A's database-which just happens to be on Node A's disks. So, how can one Node access another (dead) Node's disks? The answer is a disk subsystem that can be attached to more than one Node at a time-a feature of cluster-ready RAID subsystems. This is often referred to as Dual Attachment and has the added benefits of high performance and reliability of the disk subsystem.

The general class of...