Network Computing is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

EMC Offers SSD Cache To Boost Server Speed

12 Top Big Data Analytics Players
12 Top Big Data Analytics Players

(click image for larger view and for slideshow)

EMC unveiled on Monday a PCIe flash card with cache software to front end storage-area networks (SANs), and to accelerate server operations by keeping the 'hottest' data on high-speed single-level cell (SLC) flash memory directly on the server's PCIe bus.

VFCache, originally called Project Lightning, was discussed at last May's EMC World, and has been a source of rumor since then. Questions about whether EMC was developing its own PCIe solid state drive (SSD) and how the company could leverage this product into servers that it does not manufacture abounded.

VFCache includes the PCIe SSD and a straightforward software implementation of what is called a write-through read cache. The software intercepts read and write requests from the server's application that has been provisioned with a LUN designated for cache acceleration. When a read occurs, the driver determines if the data blocks are on the solid state drive. If so, the information is delivered to the application in less than 100 microseconds. This is called a read "hit." If the data is not on the SSD, it is read from the SAN, delivered to the application, and written to the SSD because it has just become 'hot.' This is called a read "miss." In the case of a write, the data is written to the SAN and then written to the SSD because it has just become 'hot.' In addition to caching, the SSD can be partitioned into a split-card mode, which allows a portion of the capacity to be designated for non-critical data such as logs and temporary data that does not get written to the SAN.

[ For more on server SSDS, see Which Server SSD Should You Choose? ]

This is the simplest implementation but it does have some potential for multiple servers (or virtual instances) inadvertently stepping on each other's cache data. This is prevented in the existing EMC implementation by allowing access to only a single LUN by a single server or virtual instance. EMC says it plans to eventually enhance the product to support active-active access to a single LUN from multiple servers, as well as other features to cooperate with EMC's FAST-tiered SAN arrays.

2011 was called the year of the cache because of the introduction of a number of cache products, often tied to a PCIe SSD and appearing from existing and start-up organizations. IOTurbine, for instance, launched and was purchased by PCIe manufacturer Fusion-io within months. SSD manufacturer and long-term EMC SSD supplier STEC announced a PCIe SSD of its own coupled with caching software. Start-up FlashSoft announced an SSD-agnostic cache system targeted at Windows Hyper-V and later VMware hypervisors. Silicon manufacturer LSI joined the fray in 2010 with a cache solution tied to its PCIe SSD called CacheCade and recently announced an upgraded version of the SSD. Silicon supplier PMC's Adaptec unit has been selling a version of its maxCache software with SSDs for even longer, albeit different in that it uses the RAID adapter's knowledge of the direct attached storage system to make hot data decisions, rather than agnostically accessing a SAN device.

EMC's announcement answers one question. As with other hard drives and SSDs it uses in its storage systems, the company will be purchasing finished products from OEM suppliers. Although it said multiple products had been qualified, EMC will use Micron Corporation's PCIe SSD in VFCache. Micron, no doubt, was able to compete for the business on a cost basis, as it jointly owns and operates, along with Intel, its own flash memory foundries in Utah and Singapore.

Consistent with a conservative first-product strategy, the Micron SSD announced last spring uses 34-nanometer SLC flash. Although current foundry geometries are down to 20nm and lower, the larger cell size makes access time lower and SSD use-life longer, with a penalty on capacity. Indeed, the VFCache's 300 GB capacity is relatively anemic when compared to multi-level cell (MLC) PCIe products from Fusion-io that exceed a terabyte in capacity. But a conservative approach is what you would expect from EMC.

The tougher question is how does EMC, with no server manufacturing capability, dictate to Dell, HP, IBM, and Cisco what PCIe products should be placed in their servers? The simple answer is that EMC can dictate anything it wants because of the tremendous amount of data center equipment it controls. Even Cisco, which is a close alliance partner in the V-Block VCE systems with its Unified Compute Systems, will not be able to use the initial offering due to the lack of a blade form factor in the initial product. But Micron is probably cranking out prototypes in new blade-compatible form factors as we write.

EMC hinted at future VFCache capabilities including de-duplication, intelligent dialogue with SAN controllers, active-active access collaboration between servers and virtual instances, higher capacities, and new form factors for blade support. And another future project, code-named Thunder, will produce an all-flash array of scalable, shared storage in an appliance with microsecond access times.

VFCache presently is certified with EMC storage arrays, but will work with non-EMC arrays. It is available and priced at less than other competing all-SLC PCIe SSDs.

Nominate your company for the 2012 InformationWeek 500--our 24rd annual ranking of the nation's very best business technology innovators. Deadline is April 27. Organizations with $250 million or more in revenue may apply for the 2012 InformationWeek 500 now.

Deni Connor is founding analyst for Storage Strategies NOW, an industry analyst firm that focuses on storage, virtualization, and servers.