Fault tolerant wormhole routing based on solid fault model in mesh network

Abstract

The performance of a multicomputer system is largely affected by the performance of the communication network. Nodes of a multicomputer system communicate each other by sending and receiving messages through the communication network. The performance of the communication network is determined by several factors such as topology, switching technique, flow control policy, and routing algorithm adopted in the system. It is necessary to develop an efficient routing algorithm for constructing high performance communication network. Recently, the number of nodes that compose a multicomputer system tends to increase. In a large scale multicomputer system that has thousands of nodes, the possibility of component failure is rather high. Therefore, the communication network is desirable to have ability to communicate messages despite of faulty components as well as to maintain high performance. A fault tolerant routing algorithm is mandatory to provide fault tolerant communication in multicomputer systems. A good fault tolerant routing algorithm should guarantee the correct delivery of messages under various faulty situations and exhibit high performance under both faulty and normal situation. Moreover, these goals should be achieved with moderate hardware requirement. We propose a fault tolerant routing scheme based on the solid fault model. The proposed fault tolerant routing algorithm can be applied multicomputer system constructed with mesh topology and employing the wormhole switching technique. With the solid fault model, we can cover all convex fault regions and a large portion of concave fault regions. The proposed fault tolerant routing scheme can tolerate any number of node and channel faults if the fault regions are solid fault regions. Furthermore, the fault regions are allowed to be overlapped each other and to be fault chains. To resolve non-solid concave fault regions, we also propose a fault region conversion algorithm. The fault region conversion algori...