'A reliable friend': a method for maintaining the load information in a distributed computer system

A very simple algorithm is proposed for maintaining the load information in a computer network supporting load balancing. With this algorithm, each node always has the identity of an underloaded node, if such an underloaded node exists at that moment. When there is a process that cannot be locally executed due to some time constraints, migration will only be attempted to that node. If that node cannot execute that process due to a recent change of state, the underloaded nodes will be visited one by one, until a target node is found or the process is discarded. If the state changes are not very frequent, the distributed scheduling overheads are assumed by underloaded nodes. Other advantages of the scheme are its simplicity, the graceful degradation exhibited and the low message traffic involved.<
>