Energy Optimization Based Parallel Task Scheduling Mechanism Research In Cluster Environment

The TOP500 reports show that the proportion of cluster is more and more significant in recent years, and cluster system has become the mainstream of high performance computing platform. With the development of processor and high-speed network, cluster has been widely and deeply applied to industry fields.Clusters provide a powerful computational power, as well as energy crisis. Energy crisis has appeared in high performance computing. Energy consumption has become the bottleneck in the development of cluster technology.Scheduling problem is the hot topic and focus of task scheduling research, scheduling strategy is crucial to the performance of the cluster. It has become an effective and feasible way to design a scheduling algorithm based on energy optimization.Current clustering scheduling and duplication scheduling algorithms only consider performance while completely ignoring energy consumption. Based on the consideration above, this thesis proposed a task scheduling algorithm based on energy optimization in cluster environment. Energy optimization is performed on two aspects:①This thesis constructs homogeneous cluster model, DAG(Directed Acyclic Graph) graph task model and energy consumption model, computes important parameters, and proposes parallel task scheduling algorithm based on energy optimization called EATCS(Energy-Aware Task Clustering Scheduling) in homogeneous cluster environment. And then, this thesis does some researches on the effect of processor types, network connection types and CCR (Communication -to-Computation Ratio) value on energy consumption, and compares EATCS algorithm with the past TDS(Task Duplication Scheduling) and PEBD(Performance Energy Balance task Duplication) in energy consumption in the same experiment configuration.②This thesis constructs heterogeneous cluster model, DAG graph task model and energy consumption model, computes important parameters about algorithm, and proposes parallel task scheduling algorithm based on energy optimization called EETCS (Energy-Efficient Task Clustering Scheduling) in heterogeneous cluster environment. And then this thesis simulates heterogeneous cluster environment by use of simgrid simulator, and does some researches on the effect of CCR value, node heterogeneigy and network link heterogeneigy on the energy consumption, then compares EETCS algorithm with the past TDS and NDS (Non-Duplication Scheduling) in energy consumption in the same experiment configurtion.Finally, in order to verify the effectiveness of scheduling algorithm proposed, it was applied into the Gaussian Elimination and Fast Fourier Transform classic applications.The experiment results show that the new algorithms outperform the post scheduling algorithms in energy consumption without affecting the performances.