Storage Hypervisor: Let the Contest Begin

by Administrator on June 27, 2012

This was in my email this AM, from Virsto Software CEO, Mark Davis’s PR flak, with instructions to me to use as I see fit (always a dangerous invite):

“As the industry’s first VM-centric storage hypervisor, we welcome adoption and evangelism of the storage hypervisor category by other companies such as Datacore”, said Mark Davis, CEO of Virsto Software. “It serves to further validate the need and promise of a storage hypervisor, to transform storage in the same way server hypervisor technology did for servers, providing new levels of utilization, performance, and agility. The storage industry has struggled to deliver an answer to the problem virtualization presents to traditional arrays. Virsto was pleased to launch the first storage hypervisor, and welcomes traditional storage virtualization companies like Datacore and more recently IBM SVC, to help continue to define and evangelize this exciting new category.”

This statement is worthy of Frank Luntz, I think.  Long on pleasantries, but short on the facts.

To be technically accurate, the first storage hypervisor, if you want to call it that, that I am aware of was PRISM in the IBM mainframe.  That bit of innovation, desperately needed to support mainframe multi-tenancy, let you stand up virtualized workload in Logical Partitions (LPARs) and associate with the workloads specific resources, including storage, of the mainframe sysplex.

In the x86 world, to my recollection, DataCore’s storage virtualization wares hit the market in the late 1990s, when the idea of storage virtualization was roundly condemned by the array makers who saw it as a threat to their “increase prices on rigs by adding value-add software directly to array controllers” meme.  DataCore’s story is remarkable, with the management team paying off their VCs by mortgaging their homes, key developers working for a time at no pay, transitioning sales activities to Europe where they became a household name, then transitioning back into US markets once the big players had kind-of-sort-of embraced storage virtualization again (remember this poster, making fun of the contest between IBM SVC and EMC’s burgeoning Storage Virtualization Router back in 2000?), and making a fortune.  They invented thin provisioning and extended it to all arrays, and they spearheaded efforts to go beyond mere capacity allocation and management to develop ways and means to allocate data protection services and performance optimization services as well.  Truly a remarkable tale, of which George and Ziya and Bettye should be rightfully proud.

Bottom line:  SVC from IBM came after DataCore, as did numerous other approaches from different vendors who have either disappeared or been absorbed.  With all due respect, Virsto is a newcomer to this space.  The innovators were already here and I would have to say that a real storage hypervisor is a lot more like what DataCore is offering, given that it abstracts software away from array controllers while also maintaining agnosticism regarding the platforming of server workload.  To be successful, a storage hypervisor should support ALL workloads, whether they operate from boxes running brand X server hypervisor, or brand Y, or no hypervisor at all.  DataCore does this quite successfully.  It hasn’t fallen prey to the server hypervisor holy wars, which threaten the integrity, management and investments companies have already made in storage technologies ranging from FC fabrics to NAS.

To my way of thinking, Virsto needs to be less concerned about who gets bragging rights as the first “storage hypervisor” and instead focus on what they will do when VMware announces its own proprietary storage hypervisor for its server virtualization stack, which its engineers claim will be coming soon.  That would seem to drive Virsto right out of the market.

I would part company with some storage virtualization enthusiasts in this regard.  I know that legitimate debates can be had about the right way to virtualize storage, and I know that there are many implementation approaches.

Many approaches

Many roads to Rome...

At the end of the day, however, I think we would all agree that the method we choose, whether software running in a server OS stack or pool by proxy or a hardware kit with direct connections to storage hardware via APIs, must be transparent, easily deployed, easy to manage and server hypervisor/storage hardware agnostic.  That is much more important than who coined the expression “storage hypervisor.”

I see storage virtualization merely as an enabling technology for better, more coherent and less labor intensive storage infrastructure management.  Perhaps eventually it will also provide a platform for delivering data management services, as well.

Put more simply, we need a good storage hypervisor to bend the cost curve in storage, both from the standpoint of hardware lock-ins via value-add software and from the standpoint of universal management to prevent us from having to hire more storage admins for every TB of kit we add.



{ 2 comments… read them below or add one }

is July 14, 2012 at 10:14 am

Yes, but we should also differentiate between storage virtualization and disk virtualization. It’s one thing to virtualize disks in a pool of blocks, and another to virtualize the controllers and make it perform and add value for end users. The closer you place the abstraction layer to the workflow, the better the performance, more glandular benefits, etc.

srowan July 18, 2012 at 3:26 pm

Just to clarify things, not that this is really important. Amdahl Corporation was the first to introduce a hypervisor that offered partitioning on their IBM compatible mainframes. The product was called MDF – Multiple Domain Feature.

This was followed by IBM with PR/SM – Processor Resource / System Manager.

Hitachi followed as well with Multiple Logical Partition Feature (MLPF).

All were introduced in the mid ’80′s, heady days for mainframe manufacturers.


Previous post:

Next post: