Know Your Hypervisor

Virtualization is very much on the minds of IT managers this year. Some have the vision of a low power, smoothly running, optimized data center; thinly staffed but capable of provisioning new users almost instantaneously (AKA ‘Nirvana’). Others see the nightmare of single points of (hardware) failure bringing down an entire agency, virtual machines migrating from host to host hungry for more resources, virtual server sprawl soaking up network resources exponentially, and other horrors – ending with the IT manager banished to the Smoking Tombs of Dante’s sixth circle.

The object of everyone’s attention, if not affection, is the hypervisor. The hypervisor is sandwiched between the hardware below and operating systems and application stacks above. Arguably the most critical component of a virtualization solution, the hypervisor is expected to perform transparently to the applications it serves. Like magic. Right.

The problem is that the basic Intel x86 chip architecture was never designed to be saddled with a family of virtual machines in the floors above having arguments over who gets to hold the remote control, drive the car, watch the TV, or Tivo which program. Applications that misbehave by making illegal resource calls, applications with intermittent high demand for CPU time, and other squirrely uncles must be removed from the family to achieve the virtualization vision.

Virtualization was invented by IBM in the 1960s and built-in to many high end systems from the ground up. It worked well. The objective of the current virtualization revolution, however, has been to adapt the concept to commodity hardware. Proprietary and open source hypervisors have proliferated. Think about it — why are there so many versions of a supposedly transparent layer of software to talk with a single chip architecture? Aren’t there any standards? Well … no; not yet, anyway.

Nirvana may indeed be possible but only if you prepare and execute a migration plan that minimizes risk through appropriate application grouping, load testing, and reliance on hypervisor vendors closely tied to the operating systems and application software that you will deploy.

Otherwise it may be the Smoking Tombs for you!