OpenStack Summit May 2015 Vancouver

The always growing demands for computational resources leads to our IT systems consuming already 2% of the world's energy consumption and continues to grow. Thus making data centers more energy efficient ("greener") is one of the key challenges that we face in IT. This is motivated by both economical and ecological motivation -- after all, the cost of energy has become one of the important factors of efficiently run data centers.

Virtualizing the infrastructure to enable more resource sharing and a higher resource utilization is a significant milestone in reducing the energy footprint and cost of compute infrastructure. With the success of cloud infrastructure the next level of optimization comes into focus: How do we place VMs most efficiently?

A lot of VM placement strategies were studied from a performance perspective with a focus on e.g. decreasing execution time for critical workloads or maximizing IO throughput.

The goal of this presentation is to to analyze the energy consumption of a cloud data center and discuss strategies for VM placement to optimize the energy consumption. Contrary to common beliefs, our results show that spreading (independent) workloads to multiple hosts tends to save energy as opposed to clustering it to few hosts -- at least as long as completely powering off hosts is not an option. The relationship between computational work done and energy consumption is non-linear, so placement strategies in the scheduler do make a difference also for energy efficiency