Distributed discrete-event simulations running on multi-transputer arrays

The two major techniques which are currently being used for parallel discrete-event simulation are the conservative method developed by Chandy and Misra (1971) and the optimistic technique of Jefferson (1985). In the conservative technique a process will only process its event queue if it is certain that it is safe to do so-i.e. it cannot receive an event-notification message referring to its past. Chandy and Misra (CM) developed a deadlock-free formalism for this in which every process keeps its dependants informed of the time of the earliest next message possible by sending null (non-event) messages. With the optimistic method, more commonly referred to as time warp (TW), processes disregard the possibility of event messages arriving late, but when this does occur are able to rollback and undo any erroneous work performed. Both techniques have their proponents, though neither technique is a panacea. TW has extra algorithmic complexity yet offers the capability of being quite general; on the other hand the conservative method is simpler, but is not general purpose since the connectivity of the objects must be known and built into the innards of the program. The authors describe the implementation of the TW and CM software on transputer arrays. They then deal with the performance obtained from these methods on various scales of problem. >