To have a hybrid cloud, you need a private cloud -- but does one really need to be an extension of the other?
Ask industry analysts about enterprise cloud adoption, and they’ll tell you it’s all about hybrid. Sure, startups might build entirely on the public cloud, but no large enterprise is going to move everything wholesale to AWS, Azure, or Google Cloud.
Instead, they’ll build some sort of private cloud, create new stuff on (and/or migrate some old stuff to) the public cloud, and closely integrate the two -- the definition of the hybrid cloud.
The question is, if you need a private cloud to have a hybrid cloud, where are all the private clouds? I’m not talking about merely well-managed virtualization. At a bare minimum, I mean self-service capabilities so that developers can provision their own VMs (and these days, run containers on top of them). At the high end, I mean production implementations that operate at large scale.
For self-service to work, you need automation -- or basically a library of scripts that perform common tasks: server provisioning, shared storage setup, network settings for a VLAN, and so on. Better yet, you should have orchestration assemble those automated tasks into predefined workflows for specific applications or services.
The three leading private cloud players -- Microsoft, OpenStack, and VMware -- can handle all or most of that. But who is actually implementing private clouds at scale? And what sort of path is there to a hybrid cloud, where the private and public cloud environments are parallel and can be managed of a piece?
Microsoft’s hybrid delay
Microsoft has a compelling hybrid cloud vision. The idea is that a private cloud built around Windows Server and System Center will extend to the Azure public cloud, and the two environments will be so similar you can move workloads between them seamlessly.
A considerable chunk of private cloud functionality is already in place with Microsoft’s Azure Pack, including IaaS-like virtualization management, self-service portal functionality, SQL Server (or MySQL) as a service, service management APIs, and so on. But according to Microsoft, the forthcoming Azure Stack in concert with Windows Server and System Center 2016 will be the crowning glory, delivering a private cloud very close to that of the Azure public cloud.
Last month we learned that the release of Azure Stack would be delayed until mid-2017
. Moreover, Microsoft decided that it would offer Azure Stack only through integrated systems offered by Dell, HPE, and Lenovo. The minimum hardware specs are beefy: 12 physical cores (16 optimal) and 96GB of RAM (128GB optimal). Sounds like scale-up rather than scale-out, and selling Azure Stack preloaded on hardware says something about the daunting complexity of implementing a full-fledged private cloud.
Meanwhile, Microsoft has been talking up its OMS (Operations Management Suite), a SaaS offering that enables admins to monitor and manage both private and public Microsoft cloud deployments (see Peter Bruzzese’s discussion of OMS’s security features
). Given the steep tariff for Azure Stack, my guess is that the lighter-weight OMS solution will be much more popular, leaving the full hybrid vision on hold.
OpenStack bumps along
Last week, I visited an OpenStack event in Silicon Valley to see how the leading open source private cloud was doing. The event wasn’t particularly well attended, which fed a perception that OpenStack has stalled. On the other hand, major telecoms appear to be turning to OpenStack for network function virtualization (NFV), including AT&T, Verizon, NTT Group, SK Telecom, and Deutsche Telekom. And of course there’s Walmart, which went all-in with OpenStack
a few years ago, along with (to varying degrees) American Express, Best Buy, Comcast, PayPal, andothers
Among OpenStack’s marquee customers, a common thread seems to be that they’re building customer-facing clouds at scale -- and would rather implement an open source solution than pay the operational costs to a public cloud provider. (You can imagine why Walmart wouldn’t want to run its e-commerce operation on AWS, for example.)
On the public cloud side, OpenStack has flagged. Both HP and Dell abandoned their attempts to create OpenStack public clouds, and IBM seems to be quietly backing away from the high-profile commitment it made to OpenStack
three years ago. The word “OpenStack” is conspicuously absent from IBM’s cloud home page, and aside from a few blog posts, SoftLayer -- the IaaS provider IBM bought three years ago -- barely mentions OpenStack on its site.
VMware’s cloud dreams
VMware has the biggest share of the private cloud market, depending on how you define it. Obviously, VMware still dominates virtualization. But it’s difficult to tell to what degree customers have implemented VMware’s full vRealize Suite, which includes the whole complement of monitoring, self-service, log analysis, and automation capabilities. The vRealize Business component even includes metering for chargeback.
VMware bills vRealize as a hybrid management suite for “multi-vendor, multi-cloud infrastructure.” That means you can manage across your private cloud and any public cloud that supports vSphere. Plus, VMware is pushing a public cloud offering called vCloud Air using a network of partners, but it seems to have little traction at this point.
In short, VMware is the de facto leader in private clouds (of varying capabilities), but the prospects for extension to hybrid remain unclear. Those with long memories will recall that back in 2011 Verizon bought an all-VMware public cloud play, Terremark
, to serve enterprise customers -- and like other telcos subsequently abandoned that effort. Perhaps VMware will have better luck as a result of thepartnership with IBM announced in February
, whereby SoftLayer will offer public VMware clouds as extensions to private VMware clouds. But it could be simply another flavor in IBM’s have-it-your-way cloud efforts.
What does "hybrid" mean, anyway?
Workloads run where they want to run. Some need cloud scalability and automation and self-service. Others, including legacy client-server workloads, crank along fine the way they are. Some require so much scale that you don’t want to pay a third party to host it because the operational and business dependency costs are too steep.
While I like the vision of parallel private and public clouds, where you can shift workloads around at will, workloads have a tendency to stay where you first deployed them. Containers may make that more fluid, but that portability does not necessarily demand a parallel hybrid scenario.
The bottom line is that in a modern enterprise, systems must always be integrated, whether they’re local or remote. But the hybrid idea of a private cloud as a continuous fabric extending into the public cloud is going to remain a remote possibility for most customers, mainly because the trouble and expense of building the private portion at scale remains so formidable.