基于资源预测的网格任务调度模型

跨越虚拟组织中多个域(或集群)的网格任务调度由于资源的不确定性（如动态性和异构性）而成为网格应用中亟待解决的问题。提出了一种有效的基于资源预测的网格任务调度模型——RPTS，该模型利用加权最小二乘方法进行参数估计的自回归滑动平均(ARMA)预测方法对网格环境下的主机负载进行预测。利用上述资源预测结果和一类数据并行性网格任务的建模结果,对它们进行预处理、匹配并调度执行。RPTS充分考虑了网格环境下资源的动态性和异构性,为解决网格环境下任务调度问题提供了一种较好的方法。与其他一些网格任务调度方法进行了一系列的仿真实验,结果表明RPTS模型具有任务执行时间最短和稳定性较好的特点。

Task scheduling model based on resource prediction for grid computing

Resources in grid computation environments are heterogeneous and dynamic, and tasks in grid computation environments are executed by computers from different domains or clusters of virtual organization synergistically; the static task scheduling is not fit for tasks execution in grid computing environments. In the paper, a task scheduling model based on the results of resources prediction was proposed. Firstly, a method of weighted least square estimation was given to construct Autoregressive Moving Average (ARMA) model, which would be applied in CPU load prediction of grid computer. After modeling a kind of data parallel grid tasks, the task scheduling model based on the results of resources prediction was presented. Finally the simulations on the proposed model and some other models were designed and accomplished. The simulation results demonstrate that the presented model can run both significantly faster and more stable than other models.