| SOLID-STATE DISKS (SSDs) have become a hot topic this year, beginning with their integration into EMC's Symmetrix disk array in January. Since then, dozens of companies have come out of the woodwork to say they'll manufacture or integrate SSDs.
But vendors and industry experts alike are warning that in this early market phase, the quality and capabilities of SSDs will vary widely. Some in the industry have already begun to call for standards on the manufacture or measurement of products, including some experts at the recent Flash Memory Summit 2008 in San Diego looking for SSD design standards. Michael Cornwell, manager of flash memory technology business development at Sun, says organizations can still innovate within minimal wear-leveling and performance standards. "Right now, probably every single standards body is making a grab for SSD," he says.
Because flash requires full block writes--and erasure before new writes can be committed--products need to perform an operation similar to garbage collection, a fairly well-understood process of clearing out unused memory in the DRAM world. But when it comes to doing this with the multiple flash dies inside a SATA-connected chassis that make up an SSD, there are tradeoffs to be made between durability and performance.
The average size of a flash block that must be completely erased and rewritten to modify data is 4KB to 8KB. Even if just 1KB is to be written, the entire block needs to be copied to DRAM memory, and the new bits of data must be inserted and then written again to the flash cell. Doing this quickly can require lots of writes, which limits the durability of the drive as flash tolerates a finite number of write/erase cycles. Doing it more carefully can increase durability, but slow performance, particularly when it comes to write performance, which is where some first-generation solid-state drives were even slower than hard disk drives.
Another issue affecting SSD development is the high voltage required to commit writes to the flash medium, which can lead to chip shorts and data loss. Some vendors, like Fusion-io, are shipping SSDs with RAID-on-chip technology to mitigate that effect. Other vendors have begun to experiment with the "recipe" for NAND material in their fabrication plants, while still others, like Intel, have focused on making tiny flash I/O channels as dense as possible. (Intel says its enterprise SSDs have 10 I/O channels to each die. Other vendors, such as Micron Technology, use eight.)
There's no consensus yet as to how these issues should be negotiated for different SSD applications. But, as with many other aspects of SSD technology, ask five vendors what they think about standards and you'll get five responses. Some, like Justin Sykes, director of marketing for SSD products at Micron, say standards should be created not for the design of different SSD products but for measuring how they perform in different applications.
"You could develop specs for standard usage conditions and then tie those to an endurance specification so that users can make informed choices about what apps they're buying SSDs for," says Sykes. But as for design standards, he notes, "wear-leveling is our area of differentiation. That will continue to be vendor-unique."
As one of the few storage OEMs already shipping SSDs, EMC's take, according to Symmetrix chief strategy officer Barry Burke, is that drive manufacturers sometimes rate the performance of their products according to 512-byte blocks and ignore the more realistic 4KB or 8KB block size.
"Right now, the I/O numbers on spec sheets don't tell you a lot about what the drive can really do," says Burke.