The anatomy of deadlock

In most of the classical deadlock models, the blocking rules for message-passing processes in distributed memory are considered to be the same as those for shared-variable processes in shared memory. With this approach, many obvious deadlock situations are left outside the models. We show that a distributed system can be seen not only as a composition of processes communicating by message-passing but also as a composition of processes communicating by shared variables. We call this property communication dualism. We argue that the deadlock problem should not be stated and solved in the message-passing decomposition of a distributed system, but in its shared-variable decomposition. We conclude that the emerging theory of deadlock should be based on the communication dualism property.