Storage performance is one of the biggest struggles admins of virtual server environments have to deal with, which is why so many of them gravitate toward high-performing solid-state drives. But flash storage has its own problems, such as endurance. And while all-flash arrays might provide the required performance, they are too pricey for many enterprises.
Those problems could explain the growing interest in, and expanding options around, hybrid storage arrays.
Howard Marks, founder and chief scientist at DeepStorage.net, explained to a group of IT professionals at a recent virtualization seminar that hybrid storage makes use of a small amount of flash that results in performance improvements for virtual environments. It does this in two ways, each of which automatically determines which data resides on flash, with little intervention from the administrator.
First, many hybrid arrays employ sub-LUN tiering. In this case, the controller uses information about how frequently data is accessed to determine the appropriate storage it should reside on. That information is periodically used to move hot data to faster storage and cold data to slower storage. The downside, Marks warned, is when the data movement doesn't occur frequently enough to keep up with dramatically changing workloads. To avoid the latency that might cause, Marks named several vendors that provide continuous data motion.
The alternative to sub-LUN tiering is caching, which is a real-time process. Whether a vendor is providing re-caching (accelerating reads, but not writes) or write-back caching (data is first written to the cache, then asynchronously written to disk), copies of frequently accessed data are stored on flash to allow for quicker reads. This is a good way to use the combination of flash and disk efficiently to accommodate the random workloads of virtual environments.
To hear more of what Marks had to say about hybrid storage arrays, watch the segment from his virtual seminar above.