TIME AWARE VIRTUAL MACHINE PLACEMENT AND ROUTING FOR POWER EFFICIENCY IN DATA CENTERS

Abstract

This thesis studies the problem of provisioning the resources for the specified duration of requests by selecting a set of servers for mapping their Virtual Machines (VMs) and paths for their traffic so as to maximize the acceptance rate while consuming as low power as possible. We first study the problem for flow-based time-aware requests, requiring a server-pair and an associated path for a specified duration. Second, we investigate how to exploit the VM-migration feature to save power. Third, we study the problem for time-aware application graph-based requests which characterizes the application structure and communication pattern between its components. We then investigate how to exploit the flexibility in start-time with sliding-scheduled flow-based requests to improve data center power efficiency. We finally investigate the problem for sliding-scheduled requests with application graph-based resource abstraction. For each problem, we develop optimization problem formulation, fast heuristic algorithms and carry out performance study.