New simulation techniques for energy aware cloud computing environments

Abstract

In this thesis we propose a new simulation platform specifically designed for modelling cloud computing environments, its underlying architectures, and the energy consumed by hardware devices. The models that consists on servers are divided into the five basic subsystems: processing system, memory system, network system, storage system, and the power supply unit. Each one of these subsystems has been built including new strategies to simulate energy aware. On the top of these models, there have been deployed the virtualization models to simulate the hypervisor and its scheduling policies. In addition, the cloud manager, the core of the simulation platform, is responsible for the provisioning resources management policies. It design offers to researchers APIs, allowing to perform studies on scheduling policies of cloud computing systems. This simulation platform is aimed to model existent and new designs of cloud computing architectures, with a customizable environment to configure the energy consumption of different components. The main characteristics of this platform are flexibility, allowing a wide possibility of designs; scalability to study large environments; and to provide a good compromise between accuracy and performance. A validation process of the simulation platform has been reached by comparing results from real experiments, with results from simulation executions obtained by modelling the real experiments. Therefore, to evaluate the possibility to foresee the energy consumption of a real cloud environment, an experiment of deploying a model of a real application has been studied. Finally, scalability experiments has been performed to study the behaviour of the simulation platform with large scale environments experiments. The main aim of scalability tests, is to calculate both, the amount of time and memory needed to execute large simulations, depending on the size of the environment simulated, and the availability of hardware resources to execute them.