Is 2012 gearing up to be the year of the SSD?
There's little doubt that Flash technology will continue to permeate data architectures that were once the strict purview of hard disk drives. What's intriguing for the coming year is what can be described as a "perfect storm" of developments that could push that encroachment farther and faster than many had predicted.
Exhibit A is the expected price drop of Flash to the magical number of $1 per GB. That's the threshold that memory designer Kingston Technology says will trigger widescale replacement of HDDs in desktops, servers and other systems. That trigger is expected sometime in the third quarter. Coupled with the transition to 19 nm processes and continued HDD production delays due to the recent flooding in Thailand, SSDs will likely make a compelling use case for enterprises looking to leverage high-speed networking and advanced cloud architectures.
Flash is most certainly the future of memory, according to CIOL's Shikhar Mohan Gupta, having already outclassed EEPROM in terms of cost and performance. And with its firm control of the mobile segment, Flash has a leg up when it comes to the enterprise. Still, hard disks rule the roost when it comes to capacity, and even lower-cost Flash modules will have a tough time breaking into applications that don't require high throughput.
Flash may be fast, but the problem with most SSD designs is that they still use the same SAS or SATA interface that hard disks use, according to Fusion-io. That means a 100k IOPS device can be cut down to 200 IOPS or less. Much better, says the company's Nigel Poulton, to connect directly to the PCIe bus, like Fusion-io did recently at HP's Discovery gathering in Europe earlier this month. The company showed a ProLiant server with a 2.5-inch SSD running straight to PCIe through the backplane, a setup that kicked out more than 95,000 IOPS. Rates could go even higher with broader adoption of the SCSI Express standard.
It's also true that not all SSDs are the same, especially when it comes to the critical factor of endurance, says Seagate's Teresa Worth. Devices intended for the data center have to deal with near-constant operations, random data patterns and complex read/writes - things that client drives can usually avoid. It would be nice if there were a crystal ball to determine which drives will burn their cells out faster, but barring that, enterprises can turn to the JEDEC Solid State Association's JESD219 spec, which sets endurance requirements for various application classes and provides a rating system for drive comparison.
Flash endurance can also be greatly influenced by its support systems. Sandforce recently introduced a new SSD processor, the SF-2481, that the company says is optimized for cloud environments. The device features additional error correction and failure analysis tools to promote longevity, as well as the company's DuraWrite system to improve data balance, lower write amplification and enhance garbage collection.
Ultimately, SSDs address one of the basic flaws of modern storage architectures as they strive to adapt to virtual environments and the cloud: speed. True, capacity is still important, but for the entirely new class of applications that live and die by their ability to traverse disparate data, SSDs will likely prove to be the only viable solution going forward.