System and multi-thread method to manage a fault tolerant computer switching cluster using a spanning tree

Abstract

A system, method and computer program to detect and recover from a communications failure in a computer network. The computer network has several nodes which include processor-based systems, input/output controllers and network controllers. Each node has a cluster adapter connected to multiple port switches through communications links. Data is transmitted through among the nodes through the communications links in the form of packets. A fabric manager module will monitor the network and detect a link failure. Upon the detection of a link failure between two switches a spanning tree partitioning module will partition the network into two trees at the point of the link failure. Thereafter, a link and switch identification module will identify a link between the two trees that can replace the failed link and has the least impact on the network. A routing table calculation algorithm module will calculate a new routing and distance table based on the identified link. The fabric manager module will then download the routing and distance table to only those switches effected by the new link selected to replace the failed link. This identification and recovery from communications link failures may be done with little overhead and without taking the network offline.