Atomic Actions Based on Distributed/Concurrent Exception Resolution

The state of art in concurrent exception handling and resolution is discussed and a brief outline of all research in this area given. Our intention is to demonstrate that exception resolution is a very useful concept which should be used to facilitate joint forward error recovery in concurrent and distributed systems. To do this, several new arguments are considered. We regard resolution as reaching an agreement among cooperating participants of an atomic action. It is provided by the underlying system, which makes it unified and less error prone, and this is important for forward error recovery, complex by nature. We classify atomic action schemes into asynchronous and synchronous ones, and resolution implementations into centralised and decentralised ones. Another issue that we believe to be very important is about introducing atomic action schemes based on exception resolution into existing concurrent and distributed languages, which usually have only local (one-process) exceptions. We outline the basic approach and demonstrate its applicability by showing how resolution can be used in Ada83, Ada95 (for both concurrent and distributed systems) and Java. A discussion of ways to make this concept more object oriented and, with the help of reflection, more flexible and useful, concludes the paper.