Server flash cache deathmatch: Bring it

Comment Servers currently get flash caches as a branded supplier's retrofit or through special deals between a server supplier and an OEM source, like Fusion-io. This can't go on and server flash cache is going to become a standard fit item.

Servers suffering disk storage I/O bottlenecking can now use PCI-e connected NAND caches as a tier of solid state memory between DRAM and the hard drives. This transforms the previous multi-millisecond disk I/Os into microsecond-class flash I/Os. Affordable multi-level cell (MLC) flash used in this way is going to revolutionise server-storage I/O.

The technology was pioneered by Fusion-io with its ioDrive technology and that company is now the leader in the product space, with blue chip OEMs and customers, and an IPO coming. Multiple competitors are bringing product to market and they include Intel, OCZ, Seagate, STEC, TMS, and Virident. EMC also has its Project Lightning in this space.

My feeling is that they are not long for this world unless they sign OEM supply agreements with the server OEMs. These OEMs are not interested in promoting anybody's brand but their own. They will need driver software to use the flash caches, if possible, and that software would come from their PCIe flash card supplier - at least until there is platform-level software in the server operating system.

Dell, HP, IBM, Oracle and other server manufacturers such as Cisco will, El Reg reckons, bring out servers fitted with a flash memory tier as a standard offering. Once the existing server population has been retrofitted up to saturation level then the only server flash memory tier business will largely be done by server OEMs alone. They will have their flash cache supply agreements and there'll only be a need for relatively few vendors supplying them, as there's only a need for relatively few DRAM vendors or hard disk drive vendors or CPU vendors or motherboard vendors currently - you get the picture.

Lightning is going to strike UCS

Let's try and work out what EMC is going to do with its Lightning PCIe flash. The obvious target server for this is Cisco's UCS server in VCE's vBlock. EMC will also, I believe, supplying Lightning PCIe NAND, branded VCACHE or something like that, to be fitted to host servers that connect to its VNX and VMAX arrays. EMC's channel would fit the stuff presumably and install the software drivers needed by the host server operating system.

EMC will have a list of supported servers from, we can imagine, Dell, HP, IBM and Oracle, with possibly Fujitsu and HDS on the list as well. All these qualifications or certifications will take time, months of time, and we might imagine that PowerPC servers from IBM will lag behind X86 server support. Ditto Itanium servers from HP and SPARC servers from Oracle.

We might further imagine that competing server NAND memory tier suppliers will explore the option of doing a deal with NetApp, HDS, BlueArc and other storage array vendors to hook up their server NAND to those suppliers' storage arrays so as to improve end-to-end server-to-storage I/O in the same way as EMC is proposing to do with Project Lightning.

Equally we can envisage server system and storage vendors, such as Dell, HP, IBM and Oracle looking to build equivalent integration links between their chosen server NAND memory tier and their storage, such as HP blade servers and 3PAR arrays, to achieve the same improved I/O result as EMC and so help to keep EMC out of their customer accounts

Adding a NAND tier of memory to servers is going to be a disruptive event with scope for new server flash entrants to establish themselves, and existing and start-up server NAND tier suppliers to face severe business stresses as the retrofit market transitions to a mainly server vendor supply market.

To coin a phrase, this is no (server) flash in the pan. ®