Hypercube Network Simulation Platform And Fault-tolerant Routing Algorithm

Hypercube network is one of the most important and attractive network topologies so far. With the development of computing model in the fault tolerance network for large science computing, it is very important to study the simulation platform ,which is a key part of this kind of study.This paper at first introduces the design goal and architecture of our simulation platform, then describes two key implementation technologies of platform expandability and parallel computing of the routing algorithms.The simulation platform is designed with the component technologies. It realized distributed computing by RMI technology. It contains the design and implement of web service layer, web interface layer, application logic layer and data layer.Some important features of our simulation platform are as follows: it introduces and creates the object of experimental processes dynamically and downloads relative implementation documents and resource documents to the client when necessary;it uses Pseudo Remote Threads techniques to improve the time efficiency;it adopts the component technologies to improve the development efficiency, to reuse the software, and to expand the function of simulation laboratory platform easier;it uses Java , so it is independent of platform, safe and strong;it uses implement methods of local interface and reflection mechanism to realize extension of simulation platform.With experimental results, we have analyzed the fault tolerance and effectiveness and efficiency of our parallel fault tolerant routing algorithm based on both locally k-subcube-connected fault tolerant model and locally 3-subcube-connected fault tolerant model with up to 25.0% and 12.5% faulty nodes.We develop highly scalable and strongly fault tolerant multicast routing algorithms on locally connected hypercube networks, with a large number of faulty nodes. We discuss the method of dividing locally k-subcube-connected n-dimensional hypercube network into 2^n-k k-subcubes which make the value of m which means the number of k-subcubes including all the destination nodes as small as possible. We also design and compare two algorithms, which select the path of destination subcubes. We present an algorithm to embed spanning trees into k-subcube, which permits the new joining nodes route in the scalable subcube.