EMC's DMX announcement earlier this year created a new tier 0 within the storage hierarchy, forcing competitive responses and lending credibility to niche players like Texas Memory Systems. Analysis immediately focused on the 30X greater price of flash relative to spinning disk (prices actually have been more like 40X in the past 30 days). But customers that really need the capability should base ROI on performance, not cost per GB. Flash dramatically cuts down response times and supports higher I/O rates than spinning disk. Where flash really shines is in accessing random data that is not cache friendly. An example would be a credit fraud application where a real time credit card transaction is checked against a list of suspect cards numbers where caching those numbers would not improve performance.
EMC indicates that early adopters of these devices include customers running applications that perform heavy quantitative analysis, financial transactions/electronic trades and quote systems. But good references are still hard to come by as DMX's with flash began general availability only recently on 3/26/08. New applications with potential include fraud detection/protection, telco billing and instant messaging. To support adoption, EMC has cut prices by about 1/3 from initial launch. Write performance and durability remain key customer concerns. Nonetheless, Wikibon initially expected only 1 in 100 customers to be candidates for these devices and this number could be substantially higher if prices come down rapidly.
What are the conditions by which array based flash will truly disrupt the current disk array business? There are several indicators customers should key on including:
- Price declines relative to spinning disk. Flash prices could drop as much as 10% per quarter whereas disk drive pricing declines at around one half that rate;
- Trends in Fibre Channel (FC) drive shipments-- to the extent prices of flash continue to fall and SAS demand picks up, interest in FC could begin to wane;
- The number of suppliers. There are approximately six flash and six spinning disk suppliers;
- The investment in technology and production by the flash industry;
- The emergence of players beyond STEC that address durability.
Action Item: Customers with a clear need and compelling ROI, based on IOPS, and to a lesser extent the energy efficiency of flash-based arrays, should evaluate internal flash devices relative to external memory based systems and determine the performance, costs, durability, maintainability and costs of these alternatives. Ultimately, to truly exploit these systems customers must develop data migration policies and architectures to support the facile movement of data between tiers.
Footnotes: