Allocation des ressources efficaces en énergie dans les environnements Cloud
Chaima Ghribi · HAL (Le Centre pour la Communication Scientifique Directe) · 2014
Cloud computing has rapidly emerged as a successful paradigm for providing IT infrastructure, resources and services on a pay-per-use basis over the past few years. As, the wider adoption of Cloud and virtualization technologies has led to the establishment of large scale data centers that consume excessive energy and have significant carbon footprints, energy efficiency is becoming increasingly important for data centers and Cloud. Today data centers energy consumption represents 3 percent of all global electricity production and is estimated to further rise in the future. This thesis presents new models and algorithms for energy efficient resource allocation in Cloud data centers. The first goal of this work is to propose, develop and evaluate optimization algorithms of resource allocation for traditional Infrastructutre as a Service (IaaS) architectures. The approach is Virtual Machine (VM) based and enables on-demand and dynamic resource scheduling while reducing power consumption of the data center. This initial objective is extended to deal with the new trends in Cloud services through a new model and optimization algorithms of energy efficient resource allocation for hybrid IaaS-PaaS Cloud providers. The solution is generic enough to support different type of virtualization technologies, enables both on-demand and advanced resource provisioning to deal with dynamic resource scheduling and fill the gap between IaaS and PaaS services and create a single continuum of services for Cloud users. Consequently, in the thesis, we first present a survey of the state of the art on energy efficient resource allocation in cloud environments. Next, we propose a bin packing based approach for energy efficient resource allocation for classical IaaS. We formulate the problem of energy efficient resource allocation as a bin-packing model and propose an exact energy aware algorithm based on integer linear program (ILP) for initial resource allocation. To deal with dynamic resource consolidation, an exact ILP algorithm for dynamic VM reallocation is also proposed. This algorithm is based on VM migration and aims at constantly optimizing energy efficiency at service departures. A heuristic method based on the best-fit algorithm has also been adapted to the problem. Finally, we present a graph-coloring based approach for energy efficient resource allocation in the hybrid IaaS-PaaS providers context. This approach relies on a new graph coloring based model that supports both VM and container virtualization and provides on-demand as well as advanced resource reservation. We propose and develop an exact Pre-coloring algorithm for initial/static resource allocation while maximizing energy efficiency. A heuristic Pre-coloring algorithm for initial resource allocation is also proposed to scale with problem size. To adapt reservations over time and improve further energy efficiency, we introduce two heuristic Re-coloring algorithms for dynamic resource reallocation. Our solutions are generic, robust and flexible and the experimental evaluation shows that both proposed approaches lead to significant energy savings while meeting the users' requirements