Fault-tolerant communication with partitioned dimension-order routers with complex faults
H.M.A. Fahmy, Salma Abdel-Kader Ghoneim, A. Pacha · 2002
In this paper we propose a routing algorithm to handle complex faults in multicomputer networks with dimension-order routers. Simple changes to router structure and routing logic are proposed. Our technique can be applied to current implementations in which a router is partitioned into multiple modules. While assuming local knowledge of faults, the new technique can handle complex faults effectively irrespective of their shape. By simulation we have applied it to a torus network and have shown that its performance in terms of bisection utilization and message latency is challenging. The obtained results show that the new technique performs with acceptable bisection utilization, and high adaptability to faults. Several case studies considered fault shapes such as L-shape, +-shape, and N-shape.