Contributed article by Raj Dutt, senior vice president of technology at Internap

2014: Rethinking the Hypervisor

As developers architect a new generation of scale-out applications designed to run and directly interact with the cloud, we're starting to see more industry discussion around hardware speeds and feeds, and another "old-is-new" resurgence around containerization - which raises questions about the future of the hypervisor.

Just a few months ago, Amazon released their C3 instances, and for the first time, they explicitly promise dedicated hyper-threads on specific Intel silicon. I see that as a good thing, regardless of how "un-cloud-like" it may be to talk about particular models of hardware. It's the closest Amazon has come to a dedicated server, but it's still virtualized.

For large, Internet-centric, scale-out users, performance consistency is critical. These users want to know and optimize for exactly what they're getting under the hood, down to the specific processor. This is why colocation, dedicated servers, and bare-metal cloud continue to be popular with this crowd: they can offer an unbeatable combination of price, performance, and consistency, all with zero virtualization tax.

Virtualization has become entrenched in the cloud vernacular. No doubt it has been a big enabler and has helped advance the state-of-the-art, but how strong a role will it play in tomorrow's clouds? I'd argue that the traditional enterprise virtualization value proposition is really irrelevant to these users.

They don't care about software and OS heterogeneity - they're all running a commodity Linux build and treating their infrastructure as code with configuration management. They don't care about server consolidation - they're scale out, not scale up, and everything is redlining. They're fully aware of Moore's law, just underserved and worried by it!

The opportunities for bare-metal cloud become even more interesting with the resurgence we're seeing around containerization (similar to jails, chroots, or zones, depending on your background). Docker is one of the most promising examples (open source, of course) on the horizon, allowing developers to create lightweight, portable, self-sufficient containers for applications and run them at scale. Instead of an entire virtual machine, it provides a virtual Linux environment. No virtualization tax and a boot time measured in a few seconds instead of a few minutes.

Docker seems like a great fit for bare-metal cloud, and offers a potentially more compelling way to deliver the entire performance of a bare metal server for a single tenant than virtualization. It also addresses things that are really important to web developers, which the hypervisor missed the mark on, including better support for rapid development, scalability, cross-cloud portability, roll-back, and version-control.

Is the hypervisor under siege? Not necessarily, but I think these trends are going to cause a rethinking of how we run and package applications in the cloud. It's not a zero sum game though. I suspect the more successful clouds will meld these things together and provide new and compelling ways for customers to run their applications.

##