SLA-based, Energy-Efficient Resource Management in Cloud Computing Systems

Cloud computing systems (e.g., hosting datacenters) have attracted a lot of attention in recent years. Utility computing, reliable data storage, and infrastructure-independent computing are example applications of such systems. Operational cost in these systems is highly dependent on the resource management algorithms used to assign virtual machines (VMs) to physical servers and possibly migrate them in case of power and thermal emergencies. Energy non-proportionality of IT devices in a datacenter, cooling system inefficiency, and power delivery network constraints should be considered by the resource management algorithms in order to minimize the energy cost as much as possible. Scalability of the resource assignment solution is one of the biggest concerns in designing these algorithms. This thesis examines the resource management problem in datacenters. First a centralized datacenter resource management is proposed, which considers service level agreements (SLAs) in VM placement in order to minimize the total operational cost of the datacenter. Second, a hierarchical SLA-based resource management structure is proposed, which considers the peak power constraints and cooling-related power consumption in addition to the scalability issue. The proposed hierarchical structure fits the hierarchical resource distribution in datacenters. The proposed structure is suitable to track and react to dynamic changes inside the datacenter to satisfy SLA constraints and avoid emergencies. Third, a load balancing algorithm to minimize the operational cost of a multi-datacenter cloud system is presented. Load balancing creates an opportunity to reduce the operational cost of the cloud system considering dynamic energy pricing and availability of green renewable power plants in a datacenter site.