HP’s Tech Days this week in Colorado Springs impressed on two levels. First, their willingness to engage with the analysts and writers tagged with the disreputable term “blogger.†Second, the quality of the strategy they outlined for a unified computing and storage strategy.
The outline: use a dense CPU and storage commodity-based hardware infrastructure with layered cluster storage software to build flexible and resilient scale-out storage.
Strategy
The combination of LeftHand networks scale-out iSCSI and IBRIX scale-out NFS gives the company an excellent foundation. They are still thinking through what IBRIX means for them but the fundamental promise of a common hardware infrastructure whose compute resources can easily be repurposed to scale-out storage or application services is compelling.
One surprise was that the team hadn’t grasped the implications of pNFS for their strategy. Parallel NFS is part of the NFS 4.1 spec and should be market ready, if the schedule holds, in the second half of next year.
pNFS should be a major plus for HP. If you’re selling massive scalable storage it helps to have clients capable of consuming massive capacity and bandwidth.
Pothole
The biggest hole in HP’s strategy is their go-to-market plan. They don’t have one.
The wholesale rejection of EMC’s Atmos by the EMC sales force shows the depth of the problem. Salesmen would rather sell a $500,000 box on 1 PO rather than $500,000 on 5 POs spread over a year.
HP’s sales force is less combative than EMCs but they can count dollars just as well. Corporate is going to have to think long and hard to develop a plan to get HP sales excited about selling unified storage.
The StorageMojo take
Kudos to HP for bringing us in without requiring non-disclosures. NDA’s carry the implicit promise of retribution for analysts who display too much independence, as EMC well knows.
HP’s strategy and architecture isn’t as sexy as the Atmos strategy but it may be better aligned with the sweet spot of enterprise needs. The cutting edge computer science in Atmos offers great promise, but in LeftHand and IBRIX HP has much more mature software than Atmos will have for years.
If HP aligns their sales force with the scale-out commodity hardware strategy it will force EMCs sales people to take Atmos seriously. It is a new day in enterprise storage when customers have a choice of scalable commodity-based storage systems from two major vendors.
In the larger scheme of things both HP and EMC are well-positioned. It is IBM, Hitachi and NetApp that need to sharpen their games for an even more competitive storage market.
Courteous comments welcome, of course. Disclosure: HP picked up the travel and lodging tab and I’ve done work for IBRIX in the past.
As a reminder, Robin, HP has Ibrix, Lefthand *and* PolyServe. This confuses me a bit, actually, as there appears to be some overlap, there.
Joe.
Joe,
PolyServe is intended for transactional environments – which it is very good at – but it is spec’d at 32 nodes max with 16 nodes a more practical number. It isn’t a scale-out cluster, which is why it is an enterprise, not cloud, product.
Robin
I wouldn’t say that iBrix or Lefthand are really “cloud” architectures either. LH was built and is targeted toward SMB, for simple to manage, deploy, and buy iSCSI SAN technology. iBrix has traditionally been targeted to HPC as a back-end storage platform, and while they “support” NFS, the true performance is gained via the I/O client, which has proved to be a support problem, since not everyone runs RHEL. I agree with your comments about PolyServe. I don’t think it is fair though to compare ATMOS to anything that HP has recently acquired. Time will tell to see how this “cloud” layer plays out, which open standards for data access are adopted, and really what the market will look like.
Steven,
As I noted the positions are still loose.
I’m busy the next couple of days, but one of the things I want to do is to put some market opportunity boundaries up as well. Currently I’m not sure how much commercial opportunity there is in the cloud sector.
Most of the players there are large and don’t need much help. Thus the opportunity moves over to the “private” clouds, but in the private HPC space a couple of hundred compute nodes is huge. So I’m thinking about how big the private storage cloud will be. As a first approximation management will be more important than scale.
More later.
Robin
Steven,
You’re right about LeftHand targeting the SMB customer, but the LeftHand software wasn’t purpose-built for the SMB customer, it just happened to be a convenient target market starting point for the technology.
HP LeftHand has many Cloud Customers today. As Robin described in his “Cloud Quadrant†blog, “the infrastructure must be self healing, load balancing, and fault and disaster tolerant.†HP LeftHand delivers on these capabilities.
One of the key technologies in pNFS is it allows clients to access storage devices directly and in parallel by providing data locality awareness at the client. Clustered systems without this capability are constrained by metadata lookup or I/O forwarding bottlenecks.
LeftHand’s patented DSM for Microsoft’s MPIO stack, for example, provides data locality awareness at the client today; delivering superior parallelism and scalability than other clustered storage offerings. PNFS implemented with IBRIX may just simply provide client data locality awareness in an industry standard way.
If you look at the underlying technology of IBRIX and LeftHand the possibilities are limitless.