Fault-tolerant wormhole routing in mesh networks with non-rectangular fault regions

We present our enhancement to the fault tolerance scheme proposed by Boppana and Chalasani [13]. Our aim is to create an universal fault tolerance scheme which guarantee message deliveries for remaining fault-free nodes for wormhole routing algorithms for high-radix, low-dimensional mesh networks. We propose a mark and reclaim strategy to enclose arbitrarily located faults in a specific shape that we defined as weakly convex. The fault rings and fault chains, consists of fault-free nodes and channels surrounding the fault regions, are used to route messages around the fault regions. Our algorithm ef-cube5 is an enhancement of the f-cube4 algorithm. It can handle overlapping fault-rings and fault-chains with five virtual channels. We run simulations to estimate the gain in computation power using our mark and reclaim strategy. The results show that for 12 random faulty nodes in a 16 x 16 mesh, an average of 3.66 faulty nodes can be reclaimed out of 5.30 nodes which. are marked faulty in order to make rectangular fault regions. Simulation results also show that our algorithm gives competitive performance to f-cube4.