Although solid-state, or flash, technology is already showing up in storage arrays, the current implementations...
aren't the same ones you might see in five years. For all of the hype, solid state is still a work in progress.
Solid-state drives (SSDs) are mostly finding homes as replacements for hard disk drives or as cache to speed performance. But these uses are only solid state's first foray into the data storage market.
Inserting solid state into traditional hard drive territory might not be the best way to take advantage of solid state's potential. "[Data storage] is really built on the legacy of rotating magnetic hard drives. We built around the advantages but also the disadvantages," says Jeff Janukowicz, research manager, hard disk drive components and solid-state disk drives at IDC. "Looking at the whole system of how you can better optimize it around solid-state drives is probably, longer term, the real benefit," he says.
Charlie Andrews, worldwide director of product marketing, IBM System Storage, agrees. "The least interesting implementation is putting flash in an existing storage system," he says. "You create a mismatch." He thinks systems built entirely around solid state are what we'll see in the future. And there's a lot of research happening now. "The consumer uses of flash are paying the bills that are driving technology investment," says Andrews.
One of IBM's recent solid-state research projects is Quicksilver, a collaboration with solid-state maker Fusion-io that Andrews says achieved 1 million IOPS. Researchers worked to build an optimized solid-state controller based on IBM's SAN Volume Controller (SVC) technology, and put Fusion-io's flash technology on the PCI Express bus in IBM's SVC to get the results. Andrews says the testing used a mixed workload of reads and writes, with no cache. (SVC is a block storage virtualization appliance.)
"Quicksilver was a rack-and-a-half of equipment capable of doing four times the world's record for the largest number of IOPS," says David Flynn, CTO at Fusion-io. Both companies expect the results of the project to be released as a product in the near future.
Flynn says that traditional disk arrays aren't where SSDs should go. "[Disk array vendors] are putting solid-state drives in back of their infrastructure, trying to justify that that infrastructure is providing some value," he says, when those arrays are instead handicapping the SSDs. Flynn thinks PCI Express is the best way to integrate Fusion-io's product into existing storage infrastructure.
With research, companies like IBM "are finding that NAND flash solid-state technology can be used above and beyond the ways it has been used in the past," says IDC's Janukowicz.
IBM has related projects in the works. Racetrack Memory, says the firm's Andrews, is a magnetic-based technology that could replace enterprise flash, but isn't close to commercialization. IBM is also working on phase-change memory, a thermally driven technology with symmetric read/write behavior. "You don't want to necessarily settle on one technology," says Andrews. "So we're working on a couple of them."
Chief on a wishlist of innovations, says Andrews, is making enterprise flash multilevel, and therefore twice as dense as single-level cell (SLC) solid state. Consumer flash products are multilevel, but so far only single level can handle enterprise workloads reliably.
"Highly dense, multilevel flash with enterprise-class reliability is really the invention we need to get to," says Andrews. "I haven't seen anything that would indicate yet that there's a roadmap." It's easy to set timelines for increasing density, he says, "but where are the technical underpinnings?"
Jieming Zhu, distinguished technologist at HP, agrees that multi-level cell (MLC) flash is the industry's next goal. "Enterprise solid-state drives will be based on MLC," he says. Zhu names possible tools that would let MLC flash become industry-ready, such as overprovisioning MLC memory, using DRAM, including RAID technology on the chips, and using process-level technology. "There are companies dealing with MLC with this kind of technology," he says. But none of them offer products today.
Andrews thinks there's a ways to go before users see fully developed systems built around solid state. "If you had a chart or timeline, there are parts where invention is required to get to the next step," says Andrews. "The difficult thing about innovation is that it's kind of hard to schedule."