MLC to SSD's Rescue


Solid state disks (SSDS) promise a new era in data center storage performance, offering everything from improved read and write speeds and better reliability, and a lower power footprint to boot.


Unfortunately, the great limiting factor is cost. That kind of performance comes at a premium that few enterprises are in a position to accept at the moment. So at best, SSDs are seeing limited deployments in specialty environments that depend on lightening fast throughput.


But that state of affairs my come to an end sooner than many people realize now that a number of companies are making strides in improving the performance of so-called multi-level cell (MLC) technology, bringing it closer to the more expensive single-level cell (SLC) technology that populates enterprise-class SSDs at the moment.


The difference between the two technologies has to do with the electrical charges used to store data. MLC uses several types of charge to allow more bits to be stored in a single cell. The advantage is more capacity but slower performance as the system has to negotiate its way through the various levels to find the right data. It also tends to wear out at a much faster rate than is practical for enterprise purposes.


That's why a number of manufacturers are hoping to work around these limitations and turn MLC into a viable solution. SandForce's Thad Omura says a key development is a new type of flash controller called the SSD processor. By concentrating processing power in the controller, the hope is that increased write amplification technology and other technologies can bring MLC performance up to the level of SLC while still keeping the overall system at a reasonable price point. It also opens the possibility of adding RAID-like features across multiple die, lowering the failure rate dramatically.


That's the approach that Super Talent has taken with its new MasterDrive SX system. The company used an 8-channel controller and 128 MB of cache to back up its MLC NAND flash approach, driving throughput to 220/200 MBps read/write. That's pretty close to what the company was getting on its SLC-based UltraDrive system, but at a price point that less than $370 for a 128 GB drive.


Another approach is to tweak the SLC/MLC structure itself, as Fusion-io is doing. The company has devised what it calls single mode level cell (SMLC) with an eye toward combining SLC's technological benefits with MLC's economic ones. Planned for the next releases of the 160 GB ioDrive and 320 GB ioDrive Duo systems, the technology is said to provide comparable endurance and write performance at a cost that is "substantially lower" than current SLC drives.


Toshiba, meanwhile, is touting what it calls a 3-D NAND chip coupled with 2-bit MLC technology. The official name of the still experimental approach is Pipe-shaped Bit Cost Scalable (P-BiCS) and currently exists as a 16-layer 32 Gbit prototype built on a 60 nm process. The company hopes that by stacking more layers of flash memory, they'll be able to deliver a chip that provides more capacity in a smaller form factor. Barring unforeseen difficulties, the company hopes to see a production model within three years.


Even if MLC technology brings the cost of SSDs down to their mechanical brethren, it's still unlikely that they will completely take over the data center anytime soon. As I've said before, the most likely scenario is a mixed environment in which SSDs make up the new near-line tier, reserved for that 10 percent of data that is frequently accessed and is needed at a moment's notice.


But any developments to help bring the cost of that tier to more reasonable levels is certainly welcome.