Why Hypervisors Are, And Are Not, A Commodity

My job is working on a hypervisor. Or to be technically accurate, a virtualisation platform (XenServer) that includes a hypervisor (Xen). Whatever you call it, such things have been around a long time, and many things have been built atop them, including orchestration platforms such as OpenStack and CloudStack. The idea of these, in part, is to disconnect the user from (“abstract away”) the hypervisor, and render all virtualisation platforms equal. Otherwise known as commoditisation.

If all a hypervisor does is to allow the creation of a virtual machine or two, then clearly there isn’t much difference between them. Comparing, for example, KVM and Xen, isn’t really that helpful an exercise: creating VMs isn’t that useful to you; being able to manage (everything from configuring virtual networks to live-migration) them is.

But surely the promise of OpenStack (in particular) is that it doesn’t matter which hypervisor you use? The orchestration layer makes it easy to use any (and indeed a mix of) hypervisor(s). The logical progression is that we should all choose the cheapest hypervisor, because it doesn’t matter.

Orchestration platforms are great, but…

Orchestration platforms are great, but if you want to have heterogeneous environments, you’re left with the lowest common denominator as regards features. Which is when we arrive at the point that not all virtualisation platforms are created equal.

VMware have a very large share of the server virtualisation market, certainly in the West. There’s a good reason for that: vSphere has a very wide range of features, is a well-rounded product, and also has a comprehensive set of management tools. For example, if you want lock-step fault tolerance, vSphere is pretty much the only game in town.

In contrast, many KVM-based offerings are poor cousins. That’s not to say that they’re “bad”: just that they don’t provide anything like the breadth of features or management tools (head over to WhatMatrix for a good comparison of features if you want to get a better idea).

“But wait,” I hear you say, “KVM is used for all of these hot Internet businesses!”.

“But wait,” I hear you say, “KVM is used for all of these hot Internet businesses!”. And it’s true: for workloads that are architected to be truly distributed, where one (stateless) virtual machine can disappear and a load balancer transparently redirects requests to its other, identical, cousins, there are very few clever hypervisor features that you need.

(Which is probably why container technologies, such as Docker, and serverless computing frameworks, such as AWS Lambda, are increasingly popular. But that’s a topic for an entire other post.)

Let’s be honest: almost no enterprise applications are architected to be that distributed, or have a majority of components that hold no state.

However, let’s be honest: almost no enterprise applications are architected to be that distributed, or have a majority of components that hold no state. Of course, new ones might be being architected that way (might — let’s be optimistic). But given that plenty of organisations are still using Windows XP (it’s 2016!), and IE Compatibility Mode is enabled by default for intranet sites, it’s safe to say the pace of change is glacial.

Which brings us back round to the question of whether a hypervisor can be considered a commodity. I’d like to posit a slightly different argument compared to others’ positions on the subject (links at end of post), namely that “commodity” is in the eye of the beholder, or rather, the application.

For the vast majority of enterprise workloads, lowest common denominator will not do.

Yes, hypervisors (“VM.create()” and that’s it) are a commodity. If that’s all your application needs, then great: your important choices are to be made higher up the stack. But for the vast majority of enterprise workloads, virtualisation platforms offer features that are hugely important, and for which lowest common denominator will not do.

A final thought: when I attended the May 2015 OpenStack Summit in Vancouver, I was struck by the customer stories. Not the ones on the keynote stage, with large deployments, and correspondingly large development teams, but instead those in the panel sessions, most (all?) of which had opted for VMware Integrated OpenStack (VIO). Why? Because it’s a distribution of OpenStack that integrates well with vSphere, and exposes all of its unique features. Definitely no commoditisation there.

If you’re interested in similar views from elsewhere on the Internet, head to:
- If hypervisor is commodity, why is VMware still on top? (The Register)
- Is the virtualization hypervisor a commodity? (Tech Target)
- “Hypervisor is a commodity”: A deeper analysis (Virtual Red Dot)