Even when companies do make a purchase of solid state storage, they generally don't do so across the board. Instead, they move their Tier 1 apps in to all-flash storage systems, while using hybrid or disk-based storage for their other applications.
"You hear a lot about all-flash data centers, but in reality that's not how it works, says Kerns. "People move slowly, in a far more measured way."
That means that companies are likely to continue operating mixed storage environments combining spinning hard drives and enterprise SSDs even if – or perhaps when – enterprise SSDs undercut spinning disks in raw price-per-gigabyte terms.
(Kerns pointed out that price per gigabyte is rarely a relevant metric, because it only measures is the cost of storage at rest. "It certainly doesn't tell you about what the storage can do for you, otherwise we might as well all buy tape," he said.)
Enterprise SSD Price Crossover
Deduplication, compression and space savings will continue to help close the gap between spinning disk and enterprise SSD storage costs, and a price crossover is likely to happen in the next twelve months, according to Alex McDonald, vice chair of Storage Network Industry Association (SNIA) Europe. He said these factors will continue to drive down enterprise SSD capacity prices after the crossover.
He predicted that both host-based flash (using PCIe bus-attached flash acting as DAS) and all-flash arrays with built-in deduplication and compression will see strong growth over the next few years, but controversially, he is more pessimistic for hybrid arrays with both spinning and solid state drives and flash cache in the controller. "We are not seeing growth in that area – sales look like they are flat at best or declining," he said.
Why is that? McDonald believes that a few years ago most enterprises figured that all-flash arrays filled with enterprise SSDs would be reserved for high-performance workloads because of their cost, with hybrid arrays continuing to be popular in the mainstream for mixed workloads and scalable capacities.
But things have turned out rather differently, he said. In practice the deployment of all-flash arrays has gone much faster than predicted because many enterprises have swung their investment to new, next-generation data center infrastructure, and all-flash arrays are a good fit for that. Another reason is that all-flash arrays provide cost reduction through consolidation for existing data centers (as discussed earlier.)
There's also a valuable management-related benefit to all-flash arrays filled with enterprise SSDs that's attractive to many organizations: they are significantly easier to deploy and use as you don't have to configure them to mitigate the I/O limitations of HDDs. In other words, high performance comes out of the box without the need to mess about with short-stroking, striping, and other I/O enhancing tricks.
Enterprise SSDs Will Lead to Software "Gaps"
Today most enterprise SSD use is transparent to applications: they assume that they are talking to spinning-disk storage, and smarts within the controller software deal with the intricacies of working with a solid state storage medium.
But McDonald pointed out that as enterprise SSD performance increases – either using flash or other non-volatile memory storage – then the overhead from software operations, which currently accounts for about 10 percent of total I/O read latency, will become increasingly significant. He foresees a time when storage hardware itself is no longer the bottleneck, at which point we will need a new way of programming to take advantage of the enterprise SSDs of the future. "We are not filling in the software gaps," he warns. "We need to find a better way to read data: we can't think of it as disk storage."
The problem is likely to become more acute when products based on high-performance nonvolatile storage media such as Intel and Micron's 3D XPoint become more widely available.
Photo courtesy of Shutterstock.