Abstract: A method for deadlock recovery in a shared resource multiprocess message passing computer system. The processes executing in the computer system perform periodic local checkpoints and communicate via inter-process messages. Upon detection of a deadlock in the computer system, inter-process rollback dependency is analyzed in order to choose a resource to reclaim. The choice of a resource to reclaim is made such that a resource manager which manages the resource can be rolled back so as to reclaim the resource, and such that an application process which is waiting for the resource is not rolled back past the point at which deadlock was detected. Thus, upon system restart, the reclaimed resource can be provided to the waiting process, and the waiting process can execute past the deadlock point.