UPDATED: Solid-state drives (SSD), the storage devices built on special memory that does not erase when powered down, will make their way into the datacenter and enterprise systems by 2012, predicts HP. This would be a little later than some have predicted, but it’s still another sign that the transition to solid-state is likely to happen.
Even though there have been significant advancements in SSD technology to increase capacity and improve performance, SSD started out as something for laptops, or at the low end — a thumb-sized Flash drive that plugs into a USB port.
You can’t just take that technology and use it as-is in a mission-critical datacenter application, Jimmy Daley, ISS marketing manager for HP (NYSE: HPQ), told InternetNews.com. Quite a bit of work has to go into making SSD an enterprise play — work HP is undertaking with storage partners, and also being done by its competitors, too.
Chipmakers like Intel are putting a heavy focus on getting SSD enterprise-ready, as are storage vendors like Samsung and Hitachi. EMC introduced SSDs in its enterprise products back in January 2008, and Sun has taken advantage of SSD in its “Amber Road” Storage 7000 family, seeing brisk sales with its large-scale enterprise storage systems using Flash-based SSDs.
HP offers SSD drives in its ultra-dense ProLiant BL495C server blade houses, where adding a hard drive is just not feasible due to the heat they generate. But HP maintains that SSD still isn’t ready for wider types of uses, Daley said.
“Right now, our position is today’s SSDs are ready for solution-specific scenarios, specific to a density play,” he said. “But in general, we don’t see SSD as ready for the mass market as of today. We are working with all of our partners to address this. In general, true enterprise-ready solid state drives, we believe, will come later in 2009,” with the products gaining widespread adoption by 2012.
Where SSDs fail
Because SSD drives got their start in the laptop world, they are lacking in elements vital to the datacenter market, like “hot swapping” and high-performance writing. No one is going to hot swap — replacing a drive while the system is still in use — on a laptop, so that technology was never added to the firmware, Daley said. But in a datacenter, where drives go in and out of a chassis regularly, such a feature is needed.
That means adding the right capacitance to tolerate surprise removal and surprise entry of a drive into the computer, so they can be put on hot-plug trays that can be swapped into and out of a server in the same way that a regular, hot-swappable disk drive can.
The second issue is write performance. A laptop drive can live with write performance in the neighborhood of 5,400 rotations per minute, the typical performance of a laptop hard disk. But the enterprise needs something with higher performance.
“Read performance will scream no matter what solution we put together, so we’re working with our partners to make these solutions to be much more parallel in their write operations,” Daley said.
A third problem facing SSD adoption in the enterprise is capacity: There are two kinds of SSD memory types, single-level cell (SLC) and multi-level cell (MLC). An SLC has one bit of storage per cell, while MLC has two or even three bits per cell. Higher-capacity SSD drives, reaching as high as 512B, all use MLC.
However, MLC has a shorter lifespan, degrading every time data is written to the cell. With two or three bits to a cell, that means MLC degrades faster, not just two or three time as fast, but several orders of magnitude.
Krishna Chander, senior analyst for storage research with iSuppli, estimates the native lifespan of an SLC cell at around 100,000 uses, but it drops to around 10,000 uses with two bit MLC cells and as low as 1,000 uses on a three bit cell. There are algorithms to extend that lifespan, but the bottom line is that lower-capacity SLC has a much longer working life.
Getting enterprise-ready
Yet SLC provides nowhere near enough room for blade computing, since those drives are only at 32GB to 64GB capacity. Fortunately, HP estimates that capacity will double to 64GB/128GB by the end of this year, which will finally make them feasible for use.
Then, vendors can work on performance. “We have to work on storage controllers to open up the bandwidth. If we get these devices screaming, we’ve got to make sure the full I/O path can handle everything,” Daley said.
The next step for HP is to look at where a SSD drive would make sense, and then put it in HP’s universal drive carrier, Daley said. That way, the SSD could go into the various HP products, server and rack-mounted, that accept the universal drive.
For now, however, the problems facing SSD are too numerous to make it a compelling offering for many enterprise uses. “My customers don’t want to hear about wear leveling and things like that. They want to hear this drive is warranted for three years and will last in your datacenter for x amount of time,” Daley said.
If all goes according to plan, though, HP expects to have enterprise-capable, high capacity SSD drives ready by 2012. iSupply’s Chander said that timeline is “absolutely reasonable.”
“They are right on and they have got an excellent roadmap,” he said.
Chander added that he expects SSD will be a viable storage medium in the next few years, not just for HP.
“Both hard drive companies and [Flash memory] vendors who want to be in the memory market are putting a lot of money into this for research, but there are challenges,” he said.
(Update corrects the timeline of SSD releases from EMC and Sun.)