This article is more than 1 year old

Only an open standard lance can slay EMC's cache demon

Put your heads together, lads

Blocks and Files Having EMC storage arrays direct hot data into a flash bank is great, but only if you're happy using EMC's cache gear. Wouldn't it be rather nice if there was an open and widely used server flash interface that could be used by all PCIe flash cards and PCIe flash-using vendors?

Suppose QLogic adds a flash cache to its converged network adapters (CNAs) and host bus adapters (HBAs), effectively offering PCIe flash card caching to data flying across your network. Unfortunately that wouldn't mean it could compete with EMC's Project Lightning, for QLogic has no arrangement with any storage array vendor.

Nor does it mean it can take on market leader Fusion-io. QLogic doesn't have Fusion's technology expertise or software resources. What does it have?

It has experience, channels and storage supplier credibility gained from selling hundreds of thousands of server adapters to connect to storage arrays using open interfaces. It, along with Emulex and Brocade, sells HBA and CNA products in an open market... a large open market, because of this. It would good for all three suppliers and their customers if they could carry on in this fashion with storage array-coordinated PCIe flash.

Look at this from a NetApp or HDS point of view. How can they, as standalone storage array vendors, compete with EMC's Project Lightning? They would need their own PCIe flash card plus a way to interface it efficiently with their storage array to pass and receive requests and data over the network. This equals expense and a limited market opportunity.

Ideally it would good for the industry and for customers both if storage array vendors and server adapter card suppliers got together and produced an open standard for a server flash interface. The overall server flash market would then become bigger as customers are more likely to buy products that don't lock them in.

Industry players would not have to reinvent the server flash-storage array interface for each supported storage array or PCIe flash card. Innovation would flourish on top of an open standard in this area, instead of being hobbled and restricted by multiple overlapping and essentially redundant proprietary interfaces.

Who's going to do it?

The SNIA could be an umbrella organisation under which such a standard could be developed. Suppliers with an interest in it would include, firstly, storage array vendors, such as the standalone firms (Fujitsu, HDS, NetApp, Nexsan, etc) and ones that are divisions of server vendors (Dell, HP, Huawei, IBM and perhaps Oracle). It would enable standalone storage suppliers to be closely integrated with servers in an IT environment that is prizing such integration more and more.

Secondly, the three main adapter vendors – Brocade, Emulex and QLogic – would also benefit by being able to sell into an open and larger market than one featuring their proprietary interfaces, even licensed proprietary interfaces.

Furthermore, if they wanted to cache storage array traffic on their own, they would have to track blocks coming from the array and keep repeated blocks in their cache. It is not a very efficient way of caching read data; far better to do it in conjunction with the storage array and, if possible, with applications.

Thirdly, all the current and future PCIe flash card vendors would benefit too, including Intel, LSI, Micron, OCZ, SanDisk, STEC and TMS, but less so Fusion-io. It is the market leader and will possibly perceive supporting an open server flash standard as weakening its competitive position. It would probably be right but customers would prefer it to play nice with such a standard.

Will such an open server flash-storage array interface standard happen? It would be immensely helpful to its inception if Emulex and QLogic on the one hand, and HDS and NetApp on the other publicly supported it, and committed resources to make it happen. Let IBM support it as well and the thing would be unstoppable – a done deal which EMC and Fusion-io would both be likely to support. ®

More about

More about

More about

TIP US OFF

Send us news


Other stories you might like