In December of 2013, I attended a hypervisor roundtable in New York City with local industry experts and professionals. The event, which was hosted and moderated by BWG Strategy, began with a question about whether enterprises would migrate from VMware to a multiple x86 hypervisor approach. VMware remained a central topic throughout, with a number of questions raised about the impact of OpenStack and Amazon Web Services. The event ended with a question about our choice for the key hypervisor in next five years. Most felt VMware would still remain a strong player. To my surprise, quite a few people picked Xen. I voted for KVM, which was what I had chosen for my home server.
The OpenStack Effect
Quite a few questions were raised about OpenStack. Like VMware, OpenStack creates an abstraction layer above the infrastructure. OpenStack, however, is not analogous to VMware in that it is not a hypervisor. OpenStack is a cloud management solution that integrates with different hypervisors. The default hypervisor for OpenStack is KVM, which has helped KVM gain traction among enterprises exploring OpenStack. OpenStack runs on Linux and KVM comes standard with all Linux distributions. Beyond the hypervisor, OpenStack and other cloud management solutions are pushing innovation and value up the stack. The hypervisor then becomes more of a commodity, and as this happens, I expect that we will see more rapid adoption of multiple and lower cost hypervisors.
Most, if not all, of the event attendees spoke about the current dominance of VMware. Users expressed that their organizations have had good experiences with it and, perhaps most importantly, trust it. When talking about differentiating features, one participant said “enterprise trust” was a “feature” that differentiates VMware. While not literally a feature, it was fair to say that VMware has built up a lot of equity from enterprise decision makers over the past decade. A certain level of comfort will be needed for more enterprises to adopt other hypervisors.
From a performance perspective, participants still perceived VMware to be the leader. Specifics have been documented in a 2013 paper titled “A Component-Based Performance Comparison of Four Hypervisors”. The results showed that, overall, vSphere performs the better overall among a series of test compared to KVM, Hyper-V and Xen. The other hypervisors were competitive though and each one of the four hypervisors was best in at least one benchmark. There was not one perfect hypervisor and, ultimately, different workloads may be best suited to different hypervisors. The paper concluded by saying that “effectively managing hypervisors in order to match applications to the best platform is an important, yet unstudied, challenge.”
Tools and Skills
One of the roundtable participants voiced his belief that it is critical to have a strong ecosystem of tools, especially those like VMware Workstation which can be used for home learning. VMware Workstation is an excellent tool for developing one’s hypervisor management skills. I had myself evaluated using VMware or KVM for my home server. I ultimately chose KVM. VMware had good out-of-the-box support for nested virtualization, which I wanted to build a virtual server environment for exploring OpenStack. With KVM, I had to make modifications and implemented Virtual Distributed Ethernet to get the environment to work the way I wanted. While I saved some money, I spent a lot of extra time with implementation. Today, I expect that a trade off like the one I made would be fairly representative but expect improvement in the other hypervisors.
What about Amazon Web Services?
One of the final questions was about the impact of Amazon Web Services (AWS) on VMware. On one hand, VMware is building on a hypervisor foundation and extending enterprise infrastructure to cloud services. On the other hand, AWS is attracting workload that would have otherwise run on enterprise infrastructure. The AWS model has significant economy-of-scale advantages. That said, I would not jump to the conclusion that the best price-to-performance decision is to go with a service like AWS. Consider for one that many workloads may perform better on VMware and hypervisors that take advantage of hardware-assisted virtualization capabilities than they do on AWS, which is based on para-virtualized Xen. There are other factors beyond economy-of-scale, like hypervisor choice, that need to be accounted for.
In my job, I have talked to many enterprises about their virtualization strategies. In 2013, I started to hear enterprises genuinely become more interested in multiple x86 hypervisors. In past years, I heard mostly about VMware as the only hypervisor requirement. This year, I heard many enterprises talk about interest in either Hyper-V, Xen or KVM. Beyond these compute hypervisors, some enterprises also started to explore software defined strategies that leverage not only virtualized compute but virtualized storage and networking. The ultimate benefit of a software defined environment is to optimize the infrastructure resources to the workloads. Perhaps one day this will even include automation that pairs workloads with their optimal hypervisor.