Data access rates for flash-based solid-state drives (SSDs) can be as much as 30 times faster than rotating media, so why bother to use an SSD cache? The answer is latency, which exists in all storage media, and for applications that perform complex analyses on large datasets, latency can create bottlenecks even for SSDs.
SSD manufacturer SanDisk, for example, is claiming a two- to five-times improvement in application performance using its recently launched FlashSoft for VMware vShere. “It all comes down to the application’s I/O profile,” says Rich Peterson, director of marketing management for FlashSoft. “If the application has a relatively good I/O profile [I/O load, block size, access patterns], we can see significant acceleration.”
Another benefit to caching is improved efficiency in a virtualized environment. “The reduction of latency in the servers enables you to put more virtual machines on the same physical host,” says Peterson. “That is what we refer to as VM density.” With caching, Peterson says some customers are able to “three times the amount of business” on the same physical infrastructure.
Most SSD vendors offer a caching option, but FlashSoft works across all brands of SSDs, including those from STEC, Intel, Texas Memory Systems, and Violin Memory. “We’re working with some major financial service institutions and server and storage systems partners,” says Peterson. “They all said, ‘We’d like to standardize our decision on caching software, but we want flexibility going forward with respect to our hardware procurement.”
The value of being able to cache multiple flash SSD systems through s single solution “would depend upon the performance, it could depend upon how many individual devices there are, how you are managing them, on how many actual caches that you may have on a system, and there are some applications that may use an awful lot of caching,” says Tom Coughlin of data storage consultancy Coughlin Associates.
Demand for SSDs is driven by the growth in data that companies need to store and make available to applications. The fact that SSD technology is not an all or nothing proposition is also appealing—enterprises can deploy SSD alongside existing storage infrastructure without disruption. “This is probably the first time that [our customers] considered a storage technology that they can adopt incrementally,” says Peterson. “We’re talking about a few thousand dollars per server. It’s possible to target the servers where the workloads are particularly I/O constrained and implement technology there only.”
Using SSDs for caching is not without concerns, and a common one is data consistency; system administrators need to know that the data of record in the main storage system matches what the cache flow used as the data of record. “There are going to be questions about support for existing storage processes like backup,” says Peterson. “If the server restarts for some reason, will I have data inconsistency?”
Coughlin says that flash storage vendors including SanDisk are addressing these concerns. “The reason why these are concerns is because they are intrinsic to flash memory itself and therefore the things that you do both in the flash in the design of the flash chips, but also the design of the controller architecture and firmware that determines both the endurance and the how consistent the performance will be in the products,” he says.
“Administrators are looking for reliability, consistency of performance, data that is read intensive, and caching that is write intensive,” says Peterson. “When you talk about using SSDs as a cache, they say, ‘OK, now I have to have asked about mechanical guarantees, data consistency, data recoverability…just make sure that there is no possible way that I can encounter problems with the data.’
Michael Nadeau, publisher of Data Informed, can be reached at firstname.lastname@example.org. Follow him on Twitter: @menadeau. Content Producer Megan Shaw can be reached at Megan.Shaw@WISpubs.com. Follow her on Twitter: @MeganShawDI.