Fraternising through flash to fight EMC
An anti-EMC flash force
Blocks and Files Fusion-io's flash SAN software is helping Cisco, HP and NetApp fight EMC and its Thunder/Xtremio flash SAN technology.
They need Fusion-io for two reasons. Firstly, with the networked flash SAN VMware is developing with the Thunder and Xtremio technology, EMC is becoming a serious force undermining the server system vendors.
With the coming FAST automated data movement between its capacity-optimised VMAX and VNX arrays and server VFCache and networked Thunder/Xtremio flash performance arrays, EMC has everything its customers need to run application software except the actual server hardware and system software.
EMC will be able to provide a potent mix of tiered storage capacity, very fast and low latency performance storage, and hypervisor (VMware) software linking server app I/O to local flash caches, potential storage memory, and the networked performance and capacity SANs. EMC will do this before Cisco, Dell, HDS, HP, and IBM can, which will weaken their overall offerings, as well as leave NetApp in a storage array niche.
Secondly Dell, HDS, HP, and NetApp have no technology answer to the all-flash array startups, such as GreenBytes, Kaminario, NexGen, Nimbus, Pure Storage, SolidFire, TMS, Violin Memory, and WhipTail. These ground-up All-Flash Arrays (AFAs) provide more and cheaper IOPS than traditional SAN and filer arrays – partly because of thin-provisioning, deduplication, and compression – and they are gaining data management software.
Developing their own AFAs or buying startups conscious of their potential value will both be expensive. Fusion-io's ION Data Acceleration appliance, or flash SAN, can shortcut the process for them and gift them an AFA integrated with their existing products in a few months.
Because EMC owns VMware, which is going into networking via the Nicira acquisition, the EMC/Cisco relationship is weaker. The EMC/Lenovo  server development deal sees EMC intending to use Lenovo/EMC–developed servers in EMC's storage systems, not Cisco servers.
VCE, the converged system vehicle from EMC/VMware and Cisco, has not been as fast to ramp sales as the two founding organisations had hoped. And NetApp has successfully courted Cisco with its FlexPod concept, forcing EMC to respond with VSPEX.
Cisco knows converged server-storage-networking systems are coming and that servers are in a race to run as many apps (in virtual machine containers) as they can. It is working with both LSI and Fusion-io to integrate flash caches into its UCS servers, and so have a VFCache-type capability. Our belief is that it knows it can't feed those as well from external disk drive arrays as it can from external flash arrays.
It needs external flash arrays, but there is only one obvious place to get them at present: EMC's project Thunder. How much safer it would be to get some separation from EMC, instead of added closeness, and use Fusion's ION flash SAN based on UCS server hardware. Then UCS-based AFAs (ioN flash SANs) would feed data to LSI and Fusion flash-enhanced UCS application servers.
VCE could adopt Thunder AFAs, but Cisco could put its UCS-based AFAs into FlexPods and use them elsewhere and not be dependent upon EMC for AFA technology. It could also work with Microsoft's Hyper-V as a VMware counterbalance and lessen its dependence on EMC/VMware.
There is a NetApp angle which could strengthen things as well. We'll come to that in a moment.
HP is a server, storage, and network systems vendor with stronger servers and networking than storage. It has no AFA technology, although it has an all-flash configuration of its top-end 3PAR P10000. The company faces a massed assault on its performance storage offerings by the AFA start-ups and has no good answer beyond a tactical deal with Violin Memory. Enter Fusion-io.
The DL370 is an HP gen 8 ProLiant server that becomes an AFA courtesy of HP OEM'ing Fusion's ioN software and ioMemory flash cards. This is HP's answer to EMC's Thunder. HP is also adding flash cache to its gen 8 ProLIant servers, calling it Smart Cache, and giving it a VFCache-type technology. By linking this with 3PAR arrays through automated data movement software, HP has an answer to EMC's FAST.
Another brick in HP's anti-EMC wall would be to feed its DL370 from the 3PAR arrays in the same way. That gives it a stronger answer to FAST and links its 3PAR arrays, its DL370 AFA, and its Smart Cache'd servers in a combined – and stronger – offering. The ProCurve products answer Cisco's networking offer, and EMC ends up having the most converged and integrated base set of server, storage and networking resource bricks outside of Oracle.
Needing to break out of its storage array niche and preserve its relevance in a world where performance-centric data is moving closer and closer to servers, NetApp is extending its Virtual Storage Tier (VST) caching technology outside its arrays, towards servers. It has three targets in mind: server flash caches, server flash storage memory, and AFAs.
NetApp has a collaborative alliance with Fusion-io in which VST will be integrated with Fusion software and hardware, so that NetApp FAS arrays, running ONTAP, can feed ioMemory products in servers and in an AFA. But which server hardware will be used in the AFAs?
Step forward Cisco and AFA FlexPods; it's a crystal-clear opportunity giving NetApp its anti-Thunder pill. UCS servers with ioMemory flash provide an answer to EMC's VFCache and all-flash ONTAP arrays provide a response for all-flash VMAX and VNX arrays.
Extra appeal for NetApp comes with the ION AFAs using SuperMicro and any other suitable server hardware – even the HPDL370 and whatever other HP servers might be used. Dell and IBM OEM Fusion-io flash cards, and if they produce AFAs with their servers running Fusion's ioN software, that does NetApp no harm at all.
Extra, extra appeal comes from NetApp now having a response to the AFA start-ups.
Fusion's spreading tentacles
For Cisco, HP and NetApp, the Fusion-io ioMemory hardware and ION software provide a means of combatting EMC's VFCache, project Thunder/Xtremio, and FAST technologies. What's more, the Fusion-io solution should be quicker and cheaper to put in place than developing their own server flash and AFA technology or buying an AFA start-up.
It means they are beholden to Fusion-io, but that's a limited exposure, as Fusion appears content to have a component role, at least for the time being. What's not to like?
Will Dell and IBM see the same prospects and jump the same way? If Fusion-io is successful at spreading its flash tentacles through the industry, that increasing success could make vendors anxious to differentiate themselves from their competitors and persuade them to steer clear.
Fusion-io needs to disappear into the component woodwork as one way of combatting that. Or else it should have such better – and patent-protected – technology that it blows every alternative away. That means adding dedupe, compression, thin-provisioning, VAAI support, and all that kind of good stuff, amongst other things. How that pans out is something we can't foresee.
Fusion's response to the AFA startups appears to be to use the market and channel strength of its server and storage array partners to sell its technology to customers faster and wider than any individual AFA start-up can. The AFAs are now in a race to develop and strengthen their product technologies and routes to market faster than Fusion-io and its partners. They may well seek OEM deals as one way of responding.
What is becoming clear is that Cisco, HP and NetApp are fraternising through Fusion-io to fight EMC and the AFA start-ups. In the server storage flash wars, this unlikely flash force's marching song is "I/O, I/O, it's off to Fusion-io we go." ®