Storage Clusters - Tightly Coupled vs. Loosely Coupled

Clustered storage can overcome several key obstacles that storage professionals face today; capacity, storage I/O and expandability. There are two classifications of storage clusters, loosely coupled and tightly coupled and each have their place in the environment.

Tightly coupled clusters tend to be focused on primary storage, although it is available in archive and it is even used in disk to disk backup systems. In a tightly coupled cluster the data is divided up between nodes in the cluster at some level of granularity smaller than a file, typically a block or sub block level. Unlike a typical dual controller architecture where performance peaks at about half of drive capacity and then declines, a tightly coupled cluster should see improved performance as more drives and nodes are added to the environment.

The downside is that the components of the cluster are more strict, most always come from the supplier, don't allow for flexibility of mixing manufacturers or even tiers of systems from the supplier. While some will use proprietary components with special capabilities, many will use off the shelf Intel or Xyratex storage servers and then add their software plus some sort of interconnect, typically ethernet, but some will use advanced interconnects like Infiniband. 

Loosely coupled clusters are essentially stand alone dual controller systems with a global name space that reduces the management of the environment. This has the advantage of greater flexibility in the members of the cluster and that the cluster can start with just one member. Unlike a tightly coupled cluster the data is not stored across nodes in the cluster, the entirety of the file must be on a single node. It can however be mirrored to other nodes for redundancy and some storage software will know to use the secondary copy as a backup automatically. 

Loosely coupled clusters have the downside that maximum performance and capacity is limited to the performance of the node that houses the data. The performance does not scale up as nodes are added like a tightly coupled cluster does. As a result loosely coupled clusters tend to be applied where performance is important but inexpensive capacity is more important. Cloud storage being one of the more ideal candidates. 

Both styles of clustered storage can address the problems of scale, I/O and expandability and many suppliers may be able to blur the differences between the two by either driving up dual controller performance or driving down storage costs. Review both technologies and understand which one makes the most sense for what your storage demands are.