Arrays assaulted by acceleration appliances
Waiting to be lunch
Blog Pesky acceleration appliances are proliferating and saying they can get better I/O out of storage arrays than the arrays themselves. We can imagine the array vendors sighing and saying: "Here we go again. When will these cachers ever learn?"
The pitch is that store arrays are being overwhelmed by virtualised servers and there is a large I/O gap opening up with server apps delayed by inadequate arrays bottlenecking on I/O, typically read I/O.
The latest front in the get-arrays-serving-data-faster war was opened up by Avere with a sophisticated caching product, employing close coupling of different types of cache; RAM, NVRAM, NAND and fast SAS hard drives, with different kinds of I/O; read, write, short, long and so forth. It works well and Avere, founded by ex-Spinnaker people, says you can use cheap and capacious filers for the bulk data with its greyhound of an accelerator providing the pace you need in front of them.
Alacritech came at the problem from the network optimisation angle, building on its TCP/IP acceleration prowess.
Gridiron is a fancier array cacher, saying its algorithms make its cache more responsive. We haven't seen any numbers backing this up.
All these cachers rely on drive array vendor inertia and complacency for their business model over the period from start-up to independently operating business. Is that realistic?
Let's look at EMC's Project Lightning in which server-side flash is conceptually part of the storage array's resources. This is PCIe-connected flash and EMC claims that its storage arrays, front-ended by this flash and its controlling software, can pre-load the flash with anticipated reads and satisfy server read I/O requests faster because the cached data is already in the server, not waiting out in the network. We await the numbers that will prove this.
NetApp has already added Flash Cache to its array controllers and is having a fine old success selling into VMware-using accounts with such flash-accelerated FAS arrays. EMC, as well as going one step better, it appears, with Project Lightning, has also come up with the idea of all-flash VNX and VMAX configurations.
Unless EMC has really screwed up its controllers these all-NAND monsters won't need a front-end accelerating cache at all. Let Dell, HDS, HP, Oracle and NetApp catch up and the array accelerating vendors will find their business model's horizons drawing in.
Let the other commodity array vendors like Nexsan and its ilk do the same and life will get really hard. The accelerative cache people will have to add a step-change in performance or functionality to their technology to stay out in front of these accelerated arrays, get acquired or get that IPO up and running fast, otherwise they and their venture capitalists are heading for disappointment, as they become someone else's lunch. ®