Have you noticed how much technology vendors like to talk about stuff that doesn't exist? The so-called hybrid cloud is one of the utopian dreams of cloud computing, a vision of seamless interoperability between the private cloud running in your data center and resources at your disposal in one or more public clouds. A heavily marketed work in progress, the hybrid cloud promises such colorful new use cases as "cloud bursting" and "follow the sun" and "follow the moon" computing.
Before we get to seamless, though, we'll get hybrid clouds with seams. VMware, Microsoft, and open source rivals OpenStack and CloudStack are building cloud software stacks designed for both private and public deployments, giving administrators common management tools, and developers common APIs, across the different environments. Similarly, Eucalyptus has built its private cloud business on compatibility with Amazon Web Services APIs, allowing customers who have built applications on AWS to deploy them on-premise, and vice versa.
Compatibility and consistency across private and public clouds means being able to create assets -- virtual machines, provisioning templates, automation scripts -- in one place and use them in another. It also paves the way to coordinating the use of private cloud and public cloud resources. You could run an application's Web tier in the public cloud for easier and more elastic scalability, while keeping the back-end database in the private data center, for example. Or you could run Web app servers in both clouds, triggering the provisioning of additional instances in the public cloud in response to spikes in application usage -- an example of cloud bursting.
However, while a certain degree of hybrid cloud goodness is available today, more pieces must fall into place before we can think seriously of using a public cloud as a seamless extension of the data center. First, none of the leading cloud stacks, open source or proprietary, currently supports multisite management. Even when private and public clouds share the same kinds of tools, each cloud must be managed separately.
Second, migrating live workloads from one cloud to another will require network virtualization techniques -- such as Cisco's OTV, VMware's VXLAN, or Microsoft's NVGRE -- that are only just emerging. Not even considering the distance limitations imposed by WAN latency, we're sure to see many more suns and moons before virtual machines begin hopping from data center to data center and following users around the globe.
See also: