SandForce's SF-1200 and SF-2200 controllers have popularized SSD's in recent years by offering excellent performance even in lower cost drives. The company is building on that success now with a new, enterprise-oriented chip designed for cloud computing environments. The SF-2481 is designed to offer SLC levels of reliability while using cheaper MLC NAND flash.
"As the SSD market matures, architectures are being developed to satisfy the specific needs of various market segments," said Jim Handy, SSD analyst for Objective Analysis. "SandForce is wisely targeting one of the most important SSD markets by tuning a variant of its high-performance SSD Processor to the needs of the Internet data center. This should be a good deal both for SandForce and for the company's cloud computing customers."
"Storage needs in Cloud computing is exploding worldwide, and to optimize the total cost of ownership, the cloud environment needs solid-state storage that performs at the highest levels with the greatest longevity and reliability on standard MLC flash,” said Jeremy Werner, Director of Marketing for SandForce. “The SF-2481 SSD Processor was specifically designed to deliver the perfect combination of performance, endurance, and intelligence required in cloud computing data centers.”
To date, most enterprise-class SSDs have been built on SLC flash, but the one-bit-per-cell limitation that gives SLC its speed and longevity prevents its price from dropping much. By extending the degree of ECC support and adding new early warning features, SandForce is trying to boost MLC reliability up to SLC levels.
The company's spec sheets for the new SF-2481 shed very little light on exactly how SandForce has improved its ECC capabilities. Compare the SF-2400 to the SF-2200:
The amount of ECC correction remains steady at 55b/512B. Presumably the improvements are algorithmic, as the documentation claims that the "SF-2400 SSD Processors feature a superior higher-level BCH ECC algorithm with double the standard ECC correction power above the NAND flash required specification." The 2200's data sheet, in contrast, states that the controller's data protection "surpasses the leading high-performance enterprise HDDs used today."
Specs between the two parts are nearly identical, but the SF-2400 claims an unrecoverable read error rate of less than one sector per 1017
bytes read. The SF-2200's URE rate is 1016
One bit of text prominently featured in the SF-2200's literature
in the SF-2400's
is the promise of drive lifespan. The SF-2200 documentation states: "SSD manufacturers can offer their high volume, client customers enterprise class reliability and at least 5-year lifecycles." The SF-2400 makes no promises regarding drive lifespan at all.
We expect to see a considerable degree of discussion around ECC, overprovisioning, and other means of error correction/mitigation in the next 12-18 months. At the high end of the market, companies have every reason to push MLC-based products toward the enterprise. Server price points are also robust enough to allow for a higher degree of overprovisioning without sabotaging the inherent cost advantage of MLC.
Meanwhile in the consumer market, manufacturers like OCZ are moving to adopt triple-layer cell (TLC) NAND. TLC promises to reduce costs by improving flash density, but it significantly lowers the number of write cycles the flash can handle before failure. This makes ECC correction and wear leveling even more important, and manufacturers are already struggling with how to extend these metrics to push drives into lower price points without compromising reliability.