Evolutionary Game Theoretic Power Capping for Virtual Machine Placement in Clouds

This paper studies a multiobjective evolutionary game theoretic framework for application placement in clouds that support a power capping mechanism (e.g., Intel’s Runtime Average Power Limit–RAPL) for CPUs. Given the notion of power capping, power can be treated as a schedulable resource in addition to traditional resources such as CPU time share and bandwidth share. The proposed framework, called Cielo, aids cloud operators to schedule resources (e.g., power, CPU and bandwidth) to applications and place applications onto particular CPU cores in an adaptive and stable manner according to the operational conditions in a cloud, such as workload and resource availability. This paper evaluates Cielo through a theoretical analysis and simulations. It is theoretically guaranteed that Cielo allows each application to perform an evolutionarily stable deployment strategy, which is an equilibrium solution under given operational conditions. Simulation results demonstrate that Cielo allows applications to successfully leverage the notion of power cap- ping to balance their response time performance, resource utilization and power consumption.