SimpliVity: Your legacy IT stack sucks, wanna switch it for our box?
Migrating from the mess is good, but can Kempel pull it off?
What would you say about a start-up that wants to junk all the complex, multi-layered stack clutter of today's data centres and start again with one basic scale-out OmniCube building block. Nuts right? Only it's Doron Kempel and he co-founded and sold Diligent to IBM for $200m in 2008 – so we're paying attention.
The start-up is called SimpliVity, founded and headed by Kempel, and its pitch is that today's enterprise data centre has multiple and heterogeneous server, storage and networking systems with complex layers of system and application software. It is the result of piecemeal and uncoordinated developments and is a complex and inflexible mess with a virtualisation abstraction layer and public, private and hybrid cloud structures being layered on top of it.
A SimpliVity document states:
Almost all identified problems of the infrastructure can be tied to the antiquated infrastructure stack and the coarse-grained data structure at the heart of most open systems computing platforms. This data structure limits data fluidity and scalability and is blocking the promise of server virtualisation and the cloud.
Essentially, the firm wants to get businesses to stop trying to fix this with bandaid technology advances and start afresh with a system that can scale and do what needs to be done much more simply.
The product, a scale-out building block called an OmniCube, combines server, storage and network resources seen through a virtual machine (VM) lens.
An OmniCube is:
- A 2U enclosure inside which data is globally deduplicated in 4K -8K chunks, and compressed as soon as it enters the system.
- Intel X86-driven with 10 cores, 20TB usable storage capacity, and a PCIe accelerator card - two or more OmniCubes form a networked federation which scales up by adding in more local or remote OmniCubes, with the remote ones being in a public cloud if need be.
- All OmniCube policies, operations and information are on a perVM basis and an OmniCube federation is manageable by a one VM admin through a single pane of glass.
- OmniCubes have a local backup engine and replication to another federation member for offsite protection.
- Legacy servers and VMs can connect to an OmniCube federation and invoke its storage. Legacy data and VMs can be migrated to the federation.
Kempel says a legacy IT stack includes servers and VMware, a storage switch, high-availability shared storage, a backup deduping appliance, WAN optimisation, cloud gateway, SSD array and a storage caching appliance. Kempel's pitch is that you can eliminate this multi-layered horrible hierarchy with four OmniCubes that deliver comparable payload performance, functionality and protection at one-half to one-third the acquisition cost and with a two to three times improvement in operating costs.
SimpliVity has raised $18m in venture capital funding and is targeting mid-size enterprises through the value-added reseller channel. The OmniStack software could run in any commodity server with direct-access storage and also the public cloud, but without its PCIe card accelerator.
It says says it developed its OmniStack hardware-software system after 60 engineering years and the product is in beta test. You can try out a virtual OmniCube on Amazon's EC2 cloud resource. The real thing should cost less than $50,000 per cube. You can see the company and its products at VMworld in San Francisco next week from 26 to 28 August. ®
It's Already On The Market
It has been for decades--it just turned 40, as a matter of fact, so it's pretty well shaken out. It's homogenized as it only runs on one architecture, but that architecture does everything (although localized KVM processing is usually a superior approach from all angles). So I'd say they have their work cut out for them unless the potential customers maintain their long-held animosity toward the proven product lines. In other words, it will all come down to marketing again.
as dic.usa alludes to, there are already mainframes that do all this.
That said, if you have enough requirements that even one of these is reasonable, it sounds like it'd be nice. I don't know how well this company will do, because normal people are not going to rip out and replace their whole infrastructure just because the replacement is cleaner. But, for a new deployment or when it's replacement time for your current kit, it sounds like this'd clean things up nicely. Of course the devil is in the details -- some systems sound great in theory but in practice they turn out to be a bug-ridden disaster, or clunky, or slow, or whatever (see certain fiber channel cards for a prime example.)
So they sell.... what?
They seem to sell a 2U box de dedup for 50 grand.. what else do they sell ?