The Storage World Goes Xeon

Just a few years ago there was a clear division betweenstorage products, like Windows Storage Server based NAS appliances, whichwere based on industry standard server platforms, and what the steely-eyedstorage admin considered real hardware. Real hardware used custom ASICs(Application Specific Integrated Circuit), RISC (Reduced Instruction SetComputing) processor and protocol offloads and other proprietary hardware todeliver at least midrange performance and reliability.  Today the trend is to software basedarchitectures that are faster and cheaper to bring to market.

I first noticed this trend back in 2004 when we did a VTL(Virtual Tape Library) bake-off.  All butone of the products I was testing had the familiar I/O shield and PS/2connectors of a server motherboard and that one was FalconStor's softwarerunning on an intelligent  Fiber Channelswitch.

Some high end products like 3PAR's InServ, HDS's VSP and BlueArc'sTitan still rely on custom ASICs to make the data blocks go faster but more andmore storage products are really specialized software running on pretty generichardware. 

The degree to which vendors expose that their secret saucereally is software varies.  It is, ofcourse, clearest for those vendors like Nextenta, Gluster and Caringo. They willgladly sell you their ZFS based clustered file system or RAIN(Redundant Array of Independent Nodes) object store software to run onyour own servers.  It's a bit less clearfor HP's P4000 (formerly LeftHand) iSCSI server or the unified storage systemsfrom Nimble Storage, Reldata or Overland.
Given the horsepower available in a six core Westmere XEONor 12 core Opteron, vendors can do a lot in software.  After all Compellent has been running theirmidrange arrays on Xeons for years and it helped them get innovations likeautomated tiering and Live Volume out the door before any of the competition.Even sophisticated products like IBM's StorWise V7000 arebuild around Xeons and the Storage Bridge Bay architecture.  Storage Bridge Bay (SBB) is a newmulti-vendor, or dare I call it standard, architecture that houses two "server"blades in an enclosure with disk drives. 

In SuperMicro's version each server controller holds dual Nehelemprocessors and has 3 PCI-e slots while the SAS controllers on mezzanine cardsshare access to the 16 drives in the 3U chassis.

I expect to see a lot of SBB systems over the next year asit lets vendors using clustered controller software build very costeffective, single chassis systems that can still use SAS or Fibre Channelattached JBODs (Just a Bunch of Disks) for expansion.

The real kicker will come with Intel's Jasper Forestprocessors that not only includes the PCI-e controller logic to keep chipcounts, therefore total costs down. Jasper Forest processors can share PCIe bus lanes betweendual controllers so if a component on one controller fails, the othercontroller can still access the PCI-e bus.