异构计算环境下网络路由模拟任务的非线性划分

Non-linear Partition for Network Routing Simulation Task in Heterogeneous Computing Environment

为降低异构计算环境下分布式网络模拟的计算开销,提出了一种网络路由模拟任务的非线性划分方法。该方法首先通过标准实验确定异构计算环境中各计算节点的计算能力,并通过曲线拟合确定各计算节点的路由模拟任务计算开销函数。然后基于各计算节点的计算能力以及计算开销函数,进行路由模拟任务分配及路由拓扑划分,实现对路由模拟任务的非线性划分。基于PDNS的实验表明,该方法相对于传统的均匀划分最多降低路由模拟运行时间50%以上,相对于线性划分最多降低25%以上。并随着路由模拟拓扑规模的增加,该方法的计算时间开销减少量呈现显著递增趋势。

In order to reduce the computational overhead of distributed network simulation in heterogeneous computing environment, a non-linear partition for network routing simulation task was proposed. First, through the standard experiments and curve fitting, the computing ability and routing simulation task computational overhead function of each computing node were determined in the heterogeneous environment. And then based on the computing ability and computational overhead function, the routing simulation task was assigned and the routing topology was partitioned, which realized non-linear partition for the routing simulation task. Experiments based on PDNS show that, this method reduces routing simulation running time by more than 50% compared to traditional evenly partition, and reduces by more than 25% compared to the linear method. With the increase of the routing simulation topology scale, the computational overhead decrement of this method shows remarkable increasing tendency.