* When splitting data centers is a good idea
In discussions with Nemertes Research, some IT executives have mentioned that they are building storage services centers – separate, storage-only data centers. And, while myriad issues arise when you contemplate the complete physical separation of storage and compute data centers, there are obvious benefits to the idea, and so it merits broader consideration.
The main driver of such a separation is, of course, growth – growth in the amount of data the enterprise needs for daily operations, growth in the amount of data it retains “just in case” or to satisfy regulatory requirements. The need to provide more and ever more storage – 75% growth annually, on average, in our research – chews up space and drives up power consumption, which in turn drives up heat generation, which requires more cooling. (See “Power-hungry computing” for a deeper discussion.)
An important point here is that growth in storage is outstripping growth in computing – server count is going up only by about 11% per year. Growth in storage is therefore driving the rise in resource consumption faster than the growth in compute plant. The continuing combination of multi-slot, multi-core servers and more aggressive virtualization is likely to preserve or even increase that difference.
Looking at this fact alone should compel a data center architect or manager to examine the option of separating storage and compute data centers. Separated, each can be designed to allocate space, power, and cooling in a way that reflects the needs of the different services. Servers can and do run hotter per cubic meter than most storage, so a data center built solely for servers might use cooling strategies specifically for dealing with ultra-hot hot spots, such as enclosed racks with over/under cooling, or liquid cooling systems. Storage systems don’t run so hot, and room cooling might still serve in many such centers, or less intensive forms of aisle- or rack-focused cooling.
Likewise, power consumption growth curves will be different, and separation allows enterprise companies to build storage centers with higher (and more expensive) power capacity and greater capacity for growth than compute centers. They can also contemplate placing the more power-hungry storage centers in remote locations where power is less expensive, without having to move the computing centers with them.
Careful thought has to be given to performance, of course – for some applications, only in-the-room will ever be fast enough. However, much live data is used infrequently, and much more is not really live but hasn’t been moved to near-line or off-line storage. For such classes of data, the enterprise has a higher tolerance for slower access and higher latency, and is already happy with routable technologies like iSCSI, FCIP, and NAS.
More than anything else, to really wring value from separating storage and compute centers, an enterprise must understand its actual needs for storage performance for all the kinds of information it manages, and be able to classify and move data to storage that meets those needs.
With large and obvious rewards possible, and seemingly inexorable growth driving the need to look at fresh options, the idea of conquering storage growth by dividing data centers is worth serious examination.
Are you building a storage center? Have you evaluated and rejected the option? Drop a line and tell your story!
