异构计算环境下高性能网络模拟任务划分

发布时间：2018-02-10 12:12

本文关键词： 分布式网络模拟异构计算环境路由模拟任务划分计算开销　出处：《江南大学》2014年硕士论文　论文类型：学位论文

【摘要】：作为网络研究中的一种重要的实验平台，网络模拟业已成为研究网络行为、评估网络行为、分析相关网络算法的重要手段。随着计算机网络的快速发展，所要研究的网络拓扑规模的不断扩大，然而网络模拟的计算开销量也十分庞大。解决该问题的一个有效方法就是将网络模拟任务进行划分，分配到并行计算环境中，这样每个计算节点都能分担网络模拟任务，进而提高网络模拟效率，降低网络模拟时间。而采用分布式并行模拟技术所面临的主要问题就是针对已有的并行计算环境，如何对大规模网络模拟任务进行有效划分、充分利用并行计算环境中各个计算节点的计算资源，使得各个计算节点的网络模拟任务负载达到均衡，进而大大提高网络模拟效率。分布式网络模拟面临的主要问题是如何对网络模拟任务进行合理分配，进而实现分布式网络模拟环境中各个计算节点模拟任务负载均衡。而该问题又分为以下两个方面：路由计算模拟任务的负载均衡问题；网络模拟任务负载均衡问题（即路由计算模拟与数据包转发模拟均实现负载均衡）。本文首先介绍了网络模拟、分布式网络模拟、异构计算环境下网络模拟相关概念及技术，然后据此提出了异构计算环境下模拟任务划分体系框架。在该框架中，首先通过实验手段，对异构计算环境中各个计算节点的性能进行测试，并获得各个计算节点网络模拟的相关参数：计算节点的计算能力以及路由模拟任务计算开销函数。然后针对路由模拟任务负载均衡问题，本文提出了异构计算环境下路由模拟任务的非线性划分算法(Non-linear Partition method，NLP)。经过标准实验证明：与均匀划分算法、线性划分算法相比，该算法能够充分考虑异构计算环境下路由模拟任务的负载均衡问题并且有效的降低异构计算环境下路由模拟的时间开销。最后针对网络模拟任务负载均衡问题，本文提出了异构计算环境负载均衡划分算法(Load Balanced Partition in Heterogeneous Computing Environment, LBPHCE)，该算法根据各个计算节点的数据包转发能力对网络模拟任务进行初始划分，，然后根据各个计算节点的路由计算模拟开销函数，对划分结果进行微调，进而满足异构计算环境下网络模拟中路由计算模拟的负载均衡，从而满足异构计算环境中网络模拟的负载均衡。在基于PDNS的标准实验中，通过与均匀划分算法以及线性划分算法的对比，发现LBPHCE算法能够实现异构计算环境下网络模拟任务的负载均衡，有效的降低了网络模拟的时间开销。
[Abstract]:As an important experimental platform in network research, network simulation has become an important means to study network behavior, evaluate network behavior and analyze related network algorithms. The scale of network topology is expanding, but the sales of network simulation is very large. One effective way to solve this problem is to divide the network simulation tasks into parallel computing environments. In this way, each computing node can share the task of network simulation, and then improve the efficiency of network simulation and reduce the time of network simulation. The main problem of using distributed parallel simulation technology is to aim at the existing parallel computing environment. How to divide the large-scale network simulation tasks effectively and make full use of the computing resources of each computing node in the parallel computing environment, so that the network simulation task load of each computing node can be balanced, and the efficiency of network simulation can be greatly improved. The main problem of distributed network simulation is how to distribute the tasks reasonably. The problem is divided into the following two aspects: load balancing of routing computing simulation tasks; The load balancing problem of network simulation task (routing simulation and packet forwarding simulation) realizes load balancing. This paper first introduces the concepts and technologies of network simulation, distributed network simulation and network simulation under heterogeneous computing environment, and then puts forward a framework for task partitioning in heterogeneous computing environment. Firstly, the performance of each computing node in heterogeneous computing environment is tested by experimental means, and the relevant parameters of network simulation of each computing node are obtained: computing power of computing node and computing overhead function of routing simulation task. Then, aiming at the load balancing problem of routing simulation tasks, this paper proposes a nonlinear Partition method NLP algorithm for routing simulation tasks in heterogeneous computing environment. The standard experiments show that compared with uniform partition algorithm and linear partition algorithm, This algorithm can fully consider the load balancing problem of routing simulation task in heterogeneous computing environment and effectively reduce the time cost of routing simulation in heterogeneous computing environment. Finally, aiming at the problem of network simulation task load balancing, In this paper, a load balancing algorithm named load Balanced Partition in Heterogeneous Computing Environment (LBPHCEN) is proposed, which is based on the packet forwarding ability of each computing node. Then, according to the routing simulation overhead function of each computing node, the partition results are fine-tuned to meet the load balance of routing simulation in heterogeneous computing environment. In the standard experiment based on PDNS, it is compared with uniform partition algorithm and linear partition algorithm. It is found that LBPHCE algorithm can realize load balancing of network simulation tasks in heterogeneous computing environment, and effectively reduce the time cost of network simulation.
【学位授予单位】：江南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.0

【参考文献】