面向实时并发数据流的能耗优化的GPU集群可靠处理机制研究

By using of the existing computer network with popular GPU and network ,we construct large-scale and scalable parallel GPU virtual computing environment for real-time concurrency data stream processing. According to the characteristics of real-time concurrency data stream we excavate maximization of parallel processing data stream unit. Using measurable mechanism for GPU energy consumption, we explore the approach of energy consumption measurement and energy consumption prediction for these parallel program section(process data stream unit)running on GPU. According to the these energy consumption prediction model, we found the individual electrical quality of GPU, and investigate the stable measurement model of GPU and GPU cluster. We research the cooperative resource management and scheduling mechanism for real-time concurrency data streams by these GPU parallel processing approach ,these optimization models of energy consumption for GPU, and these stable measurement models. The aim of this study is that the general GPU parallel computing environment can satisfy high performance, low cost and high reliability for the demand of real-time concurrency data stream processing. It can be widely used in large-scale data stream processing field.

利用现有的计算机网络和 “普通GPU”的计算节点，构建大规模、可扩展的并行GPU虚拟计算环境；面对实时大规模并发数据流处理特征的需求，挖掘其并行处理机制的最大化；运用可度量的能耗分析手段，研究这些并行最大化的GPU核心处理程序单元(程序切片)的能耗度量和能耗预测机制；根据这些能耗度量及预测信息，测试“普通GPU”的个体电气素质，构建GPU机群的稳定性度量模型；结合这些并行计算模型、能耗优化模型、稳定性度量模型，运用协同资源调度机制探讨面向实时大规模并发数据流处理的任务调度方法和技术，使该普通GPU并行计算环境能够满足实时大规模数据流处理所需的高速性、低成本和靠可靠性。可以广泛地应用于当今物联网大规模数据流实时处理。

利用现有的计算机网络和 “普通GPU”的计算节点，构建大规模、可扩展的并行GPU虚拟 计算环境；面对实时大规模并发数据流处理特征的需求，挖掘其并行处理机制的最大化；运用 可度量的能耗分析手段，研究这些并行最大化的GPU核心处理程序单元(程序切片)的能耗度量 和能耗预测机制；根据这些能耗度量及预测信息，测试“普通GPU”的个体电气素质，构建GPU 机群的稳定性度量模型；结合这些并行计算模型、能耗优化模型、稳定性度量模型，运用协同 资源调度机制探讨面向实时大规模并发数据流处理的任务调度方法和技术，使该普通GPU并行 计算环境能够满足实时大规模数据流处理所需的高速性、低成本和靠可靠性。可以广泛地应用 于当今物联网大规模数据流实时处理。