Distributed Fault-tolerant Routing in a Hypercube

A ABSTRACT connected hypercube containing faulty nodes/links is called an injured hypercube, f whereas a hypercube without faulty components is called a healthy hypercube. To allow non-aulty nodes in an injured hypercube to communicate with one another, each hypercube node has to be equipped with enough information so as to make correct routing decisions, and 1 y c update the information whenever the network condition changes. Obviously, it will be ver ostly for each node to keep and update complete information of the entire hypercube, i although, in that case, routing decisions will always be correct. Instead, we would like to ncorporate only minimal information into each node that is necessary for correct routing decisions. We propose two fault-tolerant routing schemes for hypercube computers, each of which f i differs in the amount of information kept in each node. The tradeoff between the amount o nformation kept in each node and the routing efficiency are analyzed. The region in which all c nodes must be informed about a component failure in order to route messages successfully is alled the influence zone of the failure. We show that the influence zone of a failure depends z on the network condition in the vicinity of the failed component, and the size of the influence one is usually fairly small relative to the size of the entire hypercube. Thus, each node needs f m to be equipped with only a small amount of global information for the successful routing o essages. Any opinions, findings, and conclusions or recom endations expressed in this publication are those of the authors and do not necessarily reflect the view of the ONR. A routing decision by a node is said to be correct if it allows each message to be delivered successfully from the node to any other node.