An Agent-Based Deadlock Detection/Resolution Algorithm for the AND Model

Previous edge-chasing Deadlock Detection algorithms for Distributed Data Base systems (DDBS) have the n steps time limitation to detect a deadlock cycle of size n in the wait-for graph (WFG). This paper proposes an agent based edge-chasing algorithm to speed up the detection process. The blocked nodes in our algorithm know their predecessors and successors simultaneously, enables they find the cycle of size 2 locally and the cycle of size n in n-2 (nge2) steps. The detection agents are assigned the originator’s priority to decrease the detection overhead, and they are forwarded or discarded momentarily to avoid the false detection. Our algorithm is built on an AND model, a retreat-inform scheme is adopted to deal with the cycle overlap problem. The correctness of the algorithm is formally proven by the invariant verification technique.