Ever since NAND flash made the jump from consumer memory sticks to enterprise storage devices the idea of simply filling up a storage array with SSDs has been out there. It certainly makes sense from a logical perspective as server-side flash devices and networkable hybrid systems proliferate, and enterprise array vendors incorporate flash into each new generation. But many users still seem to have qualms about turning over their precious data to an all-flash array, prompting the question: “Is all-flash ready for the data center?”
Why flash?; in a word “performance”, obviously. Users wanted faster boot times and memory-like responsiveness from their computers and IT professionals wanted more scalable databases and denser virtual environments without the inefficiency of throwing spindles at the problem. To deliver that performance flash storage devices were first developed as drive form-factor SSDs that were used to replace some of the spinning disk drives in a computer, server or even a shared storage array.
For server-side implementations, PCIe-based flash devices could support caching software and provide more capacity than SSDs as well as lower latencies. Most recently, we’re seeing flash devices being designed in DIMM form-factors, giving users another option with potentially much higher flash capacities and even better latency. But server-side flash in most environments was a complement to a traditional storage array, because most applications still required some kind of shared storage.
In disk-based storage arrays, SSDs and PCIe cards are being used to provide a flash tier and a new class of “hybrid” arrays, flash and disk-based appliances, have come out in recent years. Both of these types of systems typically employ some kind of software intelligence to govern the placement of data on their internal flash capacity. Whether it’s caching or tiering, they must carefully manage the use of a scarce flash resource. The chances of the wrong data being on the wrong type of storage varies between vendor but regardless there is a risk involved with predicting which data will need to be on flash. This begs the question, why not make a storage system with only flash devices?
Why not all-flash?
Instead of worrying about making sure the right data is in flash, all-flash arrays eliminate that question altogether. They can provide a high-performance solution that’s easy to implement with existing applications in an existing storage infrastructure. Unlike server-side flash solutions, all-flash arrays don’t re-introduce the potential problems data centers had with direct-attached storage had when SANs first came out.
For many use cases all-flash arrays make a lot of sense. In virtual server environments where high VM density is important, a shared flash storage array is very appealing since it supports VM and storage migration and provides high availability as well. In big data analytics implementation an all-flash array has the capacity to support the larger data sets that are often required. And for environments that involve user interaction, such as web services and e-commerce, these solutions provide the storage performance that real-time application require.
If some flash is good, is all-flash better?
We’ve established the validity of flash-based storage in the enterprise and the industry is creating solutions to continually expand the capacity of those devices. So can we jump to the ‘end game’ and simply fill a storage array completely with flash cells? Based on the growing popularity of this technology the answer would appear to be “yes”. But there may still be some concern in users’ minds about eliminating disk drives completely and going with an all-flash array.
What are the concerns?
The first concern with flash as a storage medium has always been cost. Cost-per-GB of flash is, and probably always will be, more than magnetic disk drives. Flash reliability has also created concern as an all-flash array needs to manage the fact that flash cells eventually wear out. Finally, as an enterprise storage system designed to replace a traditional disk array, all-flash arrays need to have enterprise feature sets.
For information on these concerns about all-flash arrays, please join Storage Switzerland for an informative webinar on “Overcoming the Roadblocks to the All-flash Data Center”. In it, Lead Analyst George Crump and Nimbus Data founder and CEO Tom Isakovich will discuss the issues surrounding the use of all-flash arrays and help you determine if all flash arrays are ready for your data center.
Nimbus Data is a client of Storage Switzerland