Virtualized Performance: Fusion vs. VirtualBox

Been playing around a little bit with VMware’s Fusion 2.0 and Sun’s VirtualBox 2.1.2, two virtualization platforms that are available for Mac users.

I will eventually expand this brief test to include Parallels 4.0, since it seems that Parallels and Fusion are the frontrunners when it comes to virtualization on the Mac platform.

For this initial test, I wanted to understand the differences in which the virtual disks were accessed and how that impacted performance.

Sun’s VirtualBox, although a nice little application for being free, doesn’t allow for the configuration of multiple processors, the network interface binding is a little strange, and merely doesn’t seem to have the flexibility. Fusion (and Parallels for that fact) provides the capabilities to support multiple virtual processors, works great with multiple Ethernet interfaces, and provides better integration into the platform being virtualized.

With that being said, boot time was of particular interest to me. In this scenario, I measured with a stopwatch the time from when I hit the power button on the virtual machine to the time that the GUI was up and the login screen was loaded.

I’m a Windows guy by trade, so naturally I included those numbers, but in the spirit of trying something new, I also loaded Ubuntu 8.10 desktop to see how they compared.

All of the virtual machines were configured for IDE disk access (VirtualBox does offer a SATA option) for the boot time tests. Each virtual machine had the virtualization guest extensions or additions running. Boot time was measured five times each per platform, per virtualization product and then averaged.

When comparing Windows Vista running on Fusion versus Windows Vista running on VirtualBox, Fusion loads Vista 22% (almost 9 seconds) faster than VirtualBox.

Looking at a platform other than Windows, I chose to test Ubuntu, a Linux platform available online for free. In this test, Fusion won again. This time, Ubuntu loaded almost 9% (almost 3 seconds) faster than VirtualBox.

Another interesting point is that VirtualBox supports a virtualized SATA controller. This differs from the standard IDE controller as more of the work (processing) is done by the CPU instead of the controller. I thought it would be interesting to show load times between IDE and SATA.

Using Ubuntu running on VirtualBox as an example, I found that the SATA configuration actually booted slower than that of a virtual hard disk configured to use IDE. There was almost a 3 second difference between the two.

In one of my next posts I can to cover disk file write performance on the these virtualized platforms. Also look for the addition of Parallels to the mix.