Fault-Tolerant Message Routing for Multiprocessors

In this paper the problem of fault-tolerant message routing in two-dimensional meshes, with each inner node having 4 neighbors, is investigated. It is assumed that some nodes/links can be faulty, so it is necessary to route messages, using local information at each step. A new and efficient algorithm is proposed to solve this problem. This algorithm is local and consists of pre-routing and routing stages. The pre-routing algorithm is implemented off-line. The complexity of the pre-routing stage is O(tN), where N is the number of nodes in the system, and t is the number of faulty nodes. The complexity of the online routing stage (the size of the routing table stored in the local memory) is O(t). The pre-routing algorithm is performed only once, after a new fault is detected. The algorithm allows 100% of deliverable messages to be delivered in the presence of faulty nodes with no deadlocks or lifelocks. No nodes are declared unsafe. The main idea is to construct fault-free rectangular cl...

• Type: Conference
• Year: 1998
• Where: IPPS
• Authors: Lev Zakrevski, Mark G. Karpovsky
• Keywords: Algorithm, Distributed And Parallel Computing, Faulty Nodes, IPPS 1998, Routing Stage