The G1 Gaussian-Type Resource Allocation Policy for Virtualized Data Centers: The Scaling Problem and Variation of Parameters

Virtualized data centers use techniques such as resource consolidation to reduce the energy consumption and load balancing methods to improve performance. A previously proposed resource allocation policy of Gaussian type, named G1, has a behaviour in-between resource consolidation and load balancing. This policy was previously evaluated by simulation in a small data center configuration. This paper evaluates G1 for higher dimensional data centers with up to 400 hosts and 800 virtual machines, in order to establish if the policy scales with the dimension of the data center. G1 is compared with the First Fit heuristic by simulation for time-varying workloads. Metrics such as energy consumption, the mean number of active hosts, and the number of VMs migrations are calculated for different parametrizations of the score function of the G1 policy.