Lately, I’ve been hearing a lot about so-called repatriation. That is, a company launches an application in the public cloud, and it takes off -- so much so that the cost of public cloud services become painful, and the company reels the application back to its own, private cloud to save money.
This sounds a little suspect to me. For one thing, moving an application is never as easy as it seems. For another, what sort of private cloud are we talking about and how easily can it scale if you need it to?
Private clouds are not easy to set up. In fact, very few companies have done it at scale, if you define a private cloud as local infrastructure that behaves more or less like a public cloud.
True, it’s getting easier. Microsoft, for example, has its Azure Pack, and Windows Server 2016 promises a big leap toward a real private cloud. Also, packaged OpenStack solutions are getting simpler to deploy and manage -- such as those offered by CloudScaling, IBM, PistonCloud (recently bought by Cisco), and Red Hat -- as well as OpenStack appliances from the likes of Mirantis and Breqwatr.
But which, truly, is cheaper: the public or private cloud? On-premises solutions require capital investment, operations personnel, and floor space, not to mention power and cooling. Today those selling hardware/software bundles are trying to soften the blow by offering “subscriptions” (essentially leasing plans). But terminate the lease early and you’re on the hook for the balance.
Plus, the scale aspect of the private cloud has always puzzled me. Unless you buy substantial overcapacity, you simply can’t scale quickly beyond a certain point. More important, private clouds can never incorporate as wide a range of new technologies as quickly as public clouds can offer them to customers.
Does that mean the public cloud is always better? Not at all. There are systems of record in every company that have little variability, work fine, and don’t need the magic cloud wand. Maybe one day you’ll want to migrate to a SaaS version, but if they’re doing their job capably now there’s no rush.
The boom, however, is in public-facing Web and mobile applications, which are highly variable. Increasingly, the public cloud is the place for those, not only because demand may swing wildly, but also because these apps tend to have short lifecycles.
Of course Web and mobile applications need to connect to back-end systems that often stay on-premises, which is one reason I think the interoperable private/public hybrid solutions being hatched by Microsoft, IBM, and others hold real promise. Also, if plans for the federation of OpenStack clouds bear fruit, new matrices of hybrid solutions might emerge to provide even greater flexibility.
But I don’t think we’re in the midst of a retreat from the public cloud to a captive cloud in the data center, however that may be defined. Public cloud services represent the ultimate commodification of computing. When is the last time you saw that trend run in the opposite direction?