Hu's Blog

Storage systems will need to become storage computers as more functions are being driven down into them. Old storage architectures with general purpose controllers which service all these new functions—along with the normal I/O workload—will not be able to scale. New storage architectures with separate pools of processors will be required to handle these additional functions.

Storage systems were originally designed to process read and write commands to store and retrieve data from designated block addresses on disk partitions that were identified by a Logical Unit Number (LUN), associated with a controller and target ID. Later, when RAID protection was introduced, storage systems also had to do RAID parity calculations when the data was written and RAID rebuilds in the event that one of the disks in a RAID group failed. The next step was to add business continuity, which required storage systems to support in system copies, and synchronous and asynchronous replication. All this added to the workload of the processors that were originally designed to only support reading and writing of data to storage.

Today, storage systems are expected to do much more. Thin provisioning has become a standard feature for better utilization of storage capacity, along with dynamic tiering. This requires the storage space to be broken down and managed in small chunk/chunklets, or pages. The storage processors must do substantially more work to process the meta data associated with these smaller increments of storage. In addition, the size of storage systems have increased to the size of petabytes, where they were just a few megabytes when these types of storage processors were first designed. In addition, hypervisors are pushing more work down into the storage system with APIs for formatting, copying, moving, and thin provisioning virtual disks.

While some of this extra work can be handled by faster, multi-core processors, it is time to redesign storage systems in order to process these different types of workloads more efficiently. Over the past year you have seen different vendors try to solve this problem in different ways. Hitachi introduced VSP with a new enhancement to their internal switch architecture, which was developed to optimize consolidation in SAN configurations. The enhancement was to add a separate pool of global processors (VSP) to the internal switch configuration so that the front end and back end processors could concentrate on I/O, while VSP handled the general purpose functions—like thin provisioning, tiering, replication, etc. In this way, VSP can support additional functions without impacting the basic I/O performance and throughput.

The acceptance of VSP, since its announcement on September 27, 2010, has proven the need for this type of storage computer. We expect the demand for this type of architecture to continue into 2012, with other vendors to follow with their versions of storage computers.

For Hu’s other 2012 trends, visit this bit.ly bundle: http://bitly.com/vXGP2T