Conclusion : Hyper-V vs. ESXi vs. KVM vs. VirtualBox for a home network

My performance tests indicated no clear winner in CPU-intensive tasks. Hyper-V sort of won the I/O tests, but I think you could probably get any desired outcome based on how you tweak the tests, so I'd even call that a wash.

So choosing the right virtualization solution for me comes down to usability and cost.

For me, Hyper-V seemed to have an edge in usability. I have no desire to earn a PhD in virtualization, and Hyper-V seems to have wizards and online docs which made things easier for me.

ESXi has plenty of features that would appeal to a larger 24x7 production shop, but are overkill for me. Further, all these extra bells and whistles seem to get in the way of the rather simple tasks which I'd like to do. However, if you have a large-scale virtualization project, ESXi may be the way to go.

KVM is open source and off to a great start, but I found it a little too clunky and buggy for me.

VirtualBox is "more or less" free, has fine performance, and importantly is cross platform. You can run it on Windows, Mac, or Linux hosts. It's a very good option for use by an individual or small team of developers or testers.

As far as licensing costs, Windows Server for a home network is rather modest at a few hundred dollars. For ESXi, if I read their current FAQ correctly, there is no ESXi license fee for up to 32 GB of physical RAM. KVM has no license fee. If you're thinking about rolling out a very large virtualization project, you may want to work with KVM to avoid heavy license fees.

But for me, it seems to come down to usability since performance was largely comparable. Hyper-V seems more straight-forward, ESXi has more high-end features, while VirtualBox supports multiple host OSes.

Finally, other "intangibles" may affect your decision. In my case, I work with Drupal, and there seems to be a lot of tools work in the Drupal area surrounding VirtualBox (ex. Drupal QuickStart and Vagrant). So don't forget these "intangibles" when selecting a virtualization technology.


There are 5 Comments

Hyper-V "won" the IO Benchmark? Looks to me like it is broken. That kind of values is what you get when barriers are not functional. Virtualbox is notorious for having had troubles with this in the past, and Hyper-V looked even worse in your test.

Try to benchmark a sustained write without flushing at all in your guest and see what kind of throughput you get. My guess is that all hypervisors will get values in the Hyper-V range. That would make it certain that it is indeed a bug in Hyper-V.


KVM is a good project but its still young compared to other projects like virtualbox. I had issues with KVM in the past and so I'm using virtualbox for all of my projects now. I'm using it to run Smoothwall firewall Linux OS

I still have hopes for the KVM project and I'm still following the project to this day. When they get rid of some of the issues then I will give it another shot for glory.

if you have any other projects that may beat virualbox hands down I would love to hear about.

KVM has been around for years, clearly more battle tested than hyper-v, is not Microsoft centric and is the basis of many commercial cloud and virtualization environments; Eucalyptus, RHEV (Red Hat Enterprise Virtualization which totally rocks in a multi-datacenter environment, supporting live migration, snapshots and, as said, multiple datacenters, OpenStack, and others. Especially when used in conjunction with SPICE (simple protocol for independent computing environments) KVM is quite commercial grade for thin client desktops. With the later Linux kernels RHEL 6.x, Ubuntu 12x+ a Linux hypervisor can support very dense guest OSs, be they Windows or Linux. KVM is part of Linux, VBox is a product of Oracle and Hyper-V is Microsoft centric. With the virtio pv drivers KVM has near bare metal performance. RedHat advertises and documents performance in the mid 90 percent range compared with bare metal servers.