| Summary: | This thesis focuses upon the algorithms of the so-called conservative methodology for distributed discrete-event simulation. === In the first approach, a network of fully distributed processes, and severely restricted memory are assumed. In this environment, a look-ahead, deadlock-breaking algorithm called Pseudosimulation, is developed. A lower bound on memory requirements for distributed simulation is established. === In the second approach, a scenario where several processes are mapped to the same processor is investigated. The Time-of-Next-Event algorithm, created for this purpose, computes the greatest lower bound of time stamps of the events to arrive at all empty inter-process links located in the same processor. It is shown that this algorithm has the same computational complexity as the solution of the shortest path problem. Furthermore, a distributed deadlock-breaking algorithm is introduced and proven correct. In order to increase the efficiency of the second approach, and to reduce the deadlock probability, a clustering algorithm for the assignment of processes to processors is developed. Finally, a heuristic solution to control the inflow of newly generated events and to schedule processes is proposed.
|
|---|
