Efficient distributed deadlock detection and resolution using probes, tokens, and barriers

Probes and tokens are used in many deadlock detection and resolution algorithms. A deadlock is detected by propagating probes along dependency edges. When the initiator p_i of a probe receives its probe back, it knows of the existence of a deadlock. p_i then sends out a token to clean up those probes in the deadlock; cycle which, if not removed, may later lead to phantom deadlock detections. Only after the token returns to p_i is the deadlock resolved by aborting a `victim´ (usually p_i). As a result, all involved transactions remain waiting and all involved resources locked until the token returns to p_i, although the deadlock was already detected when the probe returned to p_i. This paper proposes the idea of barriers to allow the deadlock to be resolved without waiting for the token to return to p_i, thereby reducing the average deadlock persistence time considerably