Original URL: http://www.theregister.co.uk/2011/01/11/hp_eva_future/
Where is the EVA going?
The future for HP's new toy
Comment What is HP going to do with the Enterprise Virtual Array?
The El Reg strategy department reckons a speed and feed increase is coming but no more than that, because HP doesn't own the clustering EVA functionality, and the newly-acquired 3PAR product has the virtualised data centre/cloud storage on-ramps which EVA lacks.
Why do we say this? It's because of our view of recent EVA and HP history.
In June 2007 HP launched three EVA models: the 4100, 62100 and 8100, a three-product lineup that in essence survives to the present day. These products provided a virtualised storage array for the larger small and medium businesses, and enterprise mid-range storage applications needing a Fibre Channel access SAN storage facility. Between then and today much has happened to directly and indirectly influence the EVA product set.
In February 2008, HP launched the EVA 4400, a replacement for the 4100 entry-level EVA, just eight months after the 4100, 6100 and 8100 arrays were launched. It featured FATA drives, HP's enterprise version of SATA, with up to 96 drives supported, and host access via 4Gbit/s Fibre Channel.
In November 2008 HP announced its SAN Virtualisation Services Platform (SVSP) which could virtualise HP's and other SAN arrays into single SAN storage pool. It was OEM'd from LSI which had bought StoreAge and its SAN Virtualisation product. This was a step with far-reaching product development strategy implications.
HP added the 6400 and 8400 products in March 2009, thus providing today's basic EVA product set. The new products provided more capacity and solid state drive (SSD) support. The 8440 supported up to 320 drives (including 72GB SSDs). The total was 240 max with the 8100.
Both the 6100 and 8100 were given VRAID 6, RAID 6 implemented virtually, to protect against two drive failures. There was no SAS drive support and both 4 and 8Gbit/s Fibre Channel access were supported.
That's a 21-month interval between the 6100/8100 and the 6400/8400 pairs. It's now 22 months since the 6400 and 8400 were launched so we could think we're due a refresh on product cycle grounds alone, although we have it on reasonably good authority that there will be no EVA refresh announced either this month or next.
Scale-out and virtualisation
In June 2009, newly-joined HP EMEA StorageWorks VP Gary Veale said the next-generation EVA will get a scale-out, virtualised architecture with storage processor blades (X86 servers) running storage array O/S and commodity drives in storage shelves layered underneath. There would be a separate storage management SW layer.
Some of this has come to pass. In September we heard that the EVA arrays would get thin provisioning, the ability to mostly just use disk storage for written data and not have to supply capacity up front for all the allocated space. They would also get automatic LUN Migration.
At this time David Donatelli had joined HP from EMC to run HP's server, storage and networking businesses, but was prevented from running storage for a year by an EMC lawsuit. At this point the Donatelli hire indicates HP knew it had to have a co-ordinated and integrated set of server, storage and networking products, because customers wanted much more efficient virtualised date centres with the ability for them to co-operate with cloud IT resources. Converged IT stacks were becoming the name of the game.
Big news came in April 2010 that the EVA would fit into a new naming scheme as HP said the MSA entry-level arrays were by then the P2000 products, and the acquired Left Hand Networks iSCSI SAN arrays were the P4000s. The EVA design would evolve into P6000 products with X86 controllers - along the lines Veale had described - and fit into HP"s Converged Infrastructure (CI).
There would be or could be a single set of storage enclosures with the P2000, P4000 and P6000 storage controllers pulling storage capacity out as they needed. The high-end XP line was at the top of HP's storage heap, OEM'd from Hitachi and separate from this with its USP-V controller hardware and software.
June 2010 saw hugely important announcements. HP launched its in-house developed StoreOnce D2D2500 and 4312 products to provide deduplication. These were specific and separate deduplication appliances with the Sepaton Virtual Tape Library a (VTL) providing a dedicated storage resource for larger, on-line archive applications. Deduplication could run as a data services layer application in HP's CI in the future.
The LSI angle
Secondly, EVA clusters were announced for the 6400 and 8400. Clustering was enabled through a pair of Data Path Module switches (DPM) from LSI, which ran, we think, SVSP software.
There were between two and six nodes, supporting up to 1,872 1TB FATA drives and 48 SSDs, with host access by 4 and 8Gbit/s Fibre Channel and Ethernet. A cluster could grow past six nodes by incorporating a second DPM pair.
Command View SVSP was the management tool for the single pool of storage in the cluster. Thin provisioning was supported as was LUN migration between the cluster node's arrays, fulfilling, along with the scale-out virtualisation from the DPMs and SVSP, Veale's pledge from June, 2009.
At this point the EVA had two locations for EVA storage software. The array controllers ran the HP-owned EVA software itself, years and years' worth of layers of dense and intricate code, virtualising the storage in the array, with the DPMs running the separate LSI-owned SVSP code stack providing virtualisation of storage outside and across the individual EVA arrays, thin provisioning and LUN migration. This has become a vital and strongly influential attribute affecting EVA evolution.
Thirdly, Donatelli started work on the storage business and the pace quickened. HP bought 3PAR for $2.4bn in September 2010, getting itself a hardware-accelerated storage array that overlapped the middle area of the XP and EVA product space but providing a simpler-to manage array with thin provisioning, automated tiering and other attributes that made the inServe array highly attractive to storage and other service providers. HP said the EVA had two more generations in its road-map and wasn't replaced by the 3PAR gear.
HP also OEM'd the VSP from Hitachi, the USP-V replacement, but only took it in 2.5-inch drive form, calling it the P9500, emphasising the P 'n' thousand branding scheme, and pointing up the lack of such branding for the 3PAR arrays.
In October 3PAR CEO David Scott was appointed to run StorageWorks, getting all the 'Ps' under his belt alongside InServe. Around this time the Mark Hurd imbroglio was rolling enjoyably along, with that ex-HP CEO joining arch HP enemy Oracle, and HP recruiting Leo Apothekar, an ex-SAP CEO to be Hurd's replacement.
Here we are in January, 2011 - time to refresh the EVA?
Matthew Yeager is the storage and data protection practice leader for Computacenter and he has a pretty clear idea. He sees the world of business to which the EVA is applicable being accepting of an IT model with IT providing a service to other parts of the business. That's the focus of Computacenter's activities. Many customers are demanding this model of IT provision, and not just enterprise-class customers.
They want IT to be an internal service provider with easy-to-manage storage and no need to choose between capacity, performance or manageability. They want it to connect to the virtualised infrastructure, perhaps via VAPI (VMware API), and to the cloud. Also, customers want VMAX FAST-like features (automated data tiering) in their mid-range storage now.
Yeager says customers will want to keep all their structured data on site, probably for ever. But they would like to shrink their data centre by up to two-thirds by moving their unstructured data off-site to a service provider in a federated environment and do this over the next two or three years.
Yeager's view is that the heart of storage now is software. Essential features such as deduplication, thin provisioning and increasingly autonomic management are all software features. IBM, NetApp, Dell/Compellent and EMC have taken this tack in their storage development with the features implemented inside their arrays to reduce the cost, physical space take-up and management difficulty of their storage.
He highlights IBM's STORWIZE V7000 with its integration of DS5000 storage, XIV Management, and SVC controllers in a single package as exemplifying what customers want. The EVA is an excellent and credible product with good strengths but is not as well-positioned as the 3PAR product for life in a virtualised data centre and the need to auto-provision storage for complex workloads: "You put EVA into that mix and scratch your head, and say: 'Where does that fit?'"
"The thin provisioning and other (SVSP) storage features are not built into the EVA and are not really HP, being OEM'd from LSI," says Yeager. "Maybe HP is investing funds in (storage) application development with, for example, VMware and Microsoft." Then it could go to its EVA customers and say EVA has the integration features it needs.
On the other hand Yeager thinks HP's strategy might be to keep EVA going for two more generations - the phrase 'on life-support' was used - and then provide a migration to 3PAR-developed storage.
The net of this is that HP would need a major software development effort to port the features currently in the SVSP and DPM modules into the EVA software itself, this complex and much developed mass of code. Yeager's take is that this won't be done and that HP will stick with the DPM/SVSP route and update EVA, when it refreshes the line, with array speeds and feeds.
We here at El Reg think this could mean things such as adding 2.5-inch drive tray support, multi-level cell flash support, 16Gbit/s Fibre Channel host connectivity, 2TB FATA drives, increasing the number of 3.5-inch drives supported, and possibly a SAS backplane, although that would complicate the FATA disk support.
HP could port the EVA controller to an HP X86 blade server but it would still need the LSI-sourced DPM modules and SVSP software to provide the thin provisioning, LUN migration and scale-out clustering. There is the possibility that LSI has a DPM/SVSP update coming that HP would like to take advantage of.
Yeager thinks HP recognises it has to own all the significant pieces of a converged infrastructure for virtualised data centres so that it can integrate them most effectively. That was partly why it bought 3PAR. If so, then that means the LSI-sourced DPM/SVSP products and also the Hitachi-sourced P9500 are not well placed to be included in HP's longer-term storage system development.
Yeager emphasises that there is a legitimate wish amongst the EVA customer base to be told by HP how EVA will fit into their move to converged infrastructures inside virtualised data centres with coming federation to the cloud.
We're sure HP will do the right thing by its EVA customers and provide the product refreshes and information needed to maintain its positioning of EVA as a credible and cost-effective Fibre Channel-access SAN storage product but we can't see how HP can fully integrate EVA into its converged infrastructure/virtualised data centre/cloud plans without owning the functionality provided by the LSI-sourced DPM modules and SVSP software.
That is the small elephant we see in HP's huge storage strategy room. ®