Monday, May 26th, 2008
By a long shot, virtualization is the most talked about technology in IT right now. Our customers are constently asking us for advice on the best solution available, so we put together these benchmarks in order to give them proper guidance.
System Specifications:
Host and guest VM’s are Debian Lenny with 2.6.25.4 kernel using gcc 4.3.
Video drivers used:
KVM – cirrus
VBox – vboxvideo
VMWare Workstation 6.0.3 – vmware
Tests
Our first test, and probably the most common use for a VM, was compiling. For this we used Pidgin, due to its size and popularity. Not surprisingly, VMWare came out on top in this test.
Next we tested sequential reads and writes, which is where things break apart from expectation. VMWare came in last in sequential reads, while coming in first in sequential writes.
Following sequential, we tested random reads and writes. In these tests, VMWare simply slaughtered the competition. KVM and VBox both clearly have quite a bit of catching up to do in this area.
Our last test was running glxgears to see how each VM handled X. This test was a bit a shocker, as we really didn’t expect any VM to have a substantial lead. Clearly, KVM and its use of the Cirrus driver has made it ideal for VM’s running X.
Conclusion
These benchmarks probably raised more questions than it answered, so each user should evaluate which situation best fits their needs. Is VMWare worth its nearly $200US cost? If you’re mostly focused on X or sequential reads/writes, the answer is a very loud “no”. However, if you mainly work with compilations or random read/writes, it just might be. For most of our needs, we’ll be using KVM and suggesting that our clients to do the same. This is based on an overall picture of value, not just performance or price. KVM’s openness and existing implementation allows for the most flexibility and value in our eyes.