A representative and well-understood benchmark like VMmark can be used as the basis for more elaborate experiments. As a case in point, our partners at Sun have been measuring the power consumption of their Sun Fire X4450 server while running VMmark with 8 tiles (48 total VMs). They have been kind enough to share their data with me and I have graphed it in the figure below:
The first thing one notices is the roughly 1-hour ramp-up of power consumption from about 600 Watts to an average of about 830 Watts. (The max was roughly 850 Watts.) This is followed by 3 hours of steady-state usage and a sharp decline in consumption at the end. This mirrors a typical VMmark run where the workload VMs are ramped up in a staggered fashion followed by a 3-hour measurement interval before the benchmark ends. Intuitively, the power consumption of the server should rise with the increasing work being done, and it does. I find that confirmation itself quite valuable.
Much is made of the power savings potential of server consolidation using virtualization. I daresay that one would have a difficult time running 48 physical servers, no matter how efficient, on an average of 830 Watts. A simplistic analysis would show that, on average, each server could consume only 17.5 Watts. I’d like to see that server. Kudos to the folks at Sun for their excellent work in demonstrating this potential. If you want more information on the Sun Fire X4450, please check out the VMmark results page.