The SNW Report: Lots Of Flash, Cache Is King

This week I kicked off my spring trade show and conference season at Storage Network World at the Santa Clara Convention Center.  It was the first time in recent memory that the show was in Silicon Valley, the center of the tech universe, and that seemed to have brought out a crowd with more end-users than in the past.  Aside from the all encompassing cloud, I heard a lot of buzz about flash memory based products, especially those using flash as cache for other storage systems.

My conference schedule has really gotten out of hand this year.  The second week in May I'll be speaking at Interop at one end of the Las Vegas strip at Mandalay Bay and covering EMCworld at the Venetian most of the way to the other end of the strip. Add in that it will be my last visit to the soon to close forever Sahara, where for years they've comped this low roller craps player, and I'll be driving up and down the strip most of the week.

Then I have PR folks begging me to go to NAB and The Tape Summit next week, Symantec Vision and Brocade Tech Day the first day in May and several other vendor worlds.  I've reached the point where I'm turning down invitations or I'd never get any time in the lab to do paying work.

In addition to its more obvious application as solid state disk, flash memory is changing the way we think about storage caches.  I remember a few years ago sitting in an Exchange optimization session at Microsoft's TechEd, another conference I had to prune from my overwhelming schedule this year, where the speaker said that cache had no impact on Exchange performance.

Of course what he meant was that upgrading the cache on your PERC or SmartArray controller from 4MB to 16MB wouldn't matter. Today with flash based caches we're not talking about a cache that's 1/100th of a percent the size of the data set behind it, but caches of 1-15%. While 4MB in front of 100GB may not have much effect on performancem 500GB in front of 50TB sure will.Several vendors announced new caching products or significant enhancements to their existing caching appliances at or shortly before SNW.

DataRAM announced that they're doubling the size of their flash based XcelaSAN cache appliance from 256GB to 512GB.  XcelaSAN is the only external caching appliance for Fibre Channel SANs that I'm aware of.  Users with 2-3 year old SAN arrays can add a cache layer to their SANs a lot easier and cheaper than the fork lift upgrade many vendors would require to add an internal flash cache or tier.  DataRAM announced the XcelaSAN over a year ago and was quietly working to add high availability clustering, which most of their target customers would consider a must have, before making a real market push. Now that you can cluster a pair of XcelaSANs, it could be a good fit in a lot of data centers.

Like DataRAM, Avere announced new bigger, faster external NAS cache appliances at the show, on top of the even bigger announcement a few weeks before that they'd integrated a global name space into their systems. Unlike DataRAM, Avere's appliances support multiple tiers, including RAM and 15K RPM spinning disks as well as flash. In fact, the folks at Avere would rather I call them tiered storage appliances than caches.  The new FXT 2550 and 2750 use Nehalem processors and support more RAM than their predecessors.

Perhaps the most interesting new product at SNW was Marvell's Dragonfly virtual storage accelerator, a PCIe cache card for server side caching.  The card has on-board RAM and SAS/SATA ports for SSDs. It then uses the combined RAM and flash as a multilayer cache in front of any kind of storage the server may be using, including SAN and NAS.  The devil is in the details and I have lots of questions about how Marvel's software handles things like vMotion and snapshots provided by the back end storage system, but I'll take a wait and see attitude for now.