An Optimistic PDES Protocol Over Dynamic Cloud

Abstract

Parallel discrete event simulation (PDES) simply known as parallel or distributed simulation is the execution of single discrete event simulation program on a parallel computers. PDES is largely accepted to promote interoperability and reusability of simulation components and to speed up the large scale simulations. In PDES, simulation is divided into small tasks which are placed on different processor for execution. Hence, the simulation system can be viewed as the combination of concurrent processes, each of them executing in a sequential manner on the separate processor and modeling some part of the physical system. All these processes communicate with each other by sending time stamped event messages. An event can be any update in the state of the simulation system at any simulation timestamp. The events are scheduled in timestamp order. Events are causal related to each other and processed in timestamp manner. The other adopted techniques are conservative and optimistic. In conservative approach, the causality errors are not allowed and each event is processed after all the events which could affect it. In optimistic approach allow causality errors and use some detection and recovery mechanisms for the efficient processing of the events. Timewrap is the most commonly known protocol. In timewrap, whenever a causality error occurred a rollback mechanism initiated to recover the simulation to a state of correct computation. In order to roll back other processes, system sends the anti messages to other processes whenever a causality error occurs so that all the events that are processed prematurely can be undone. Thus, Roll back mechanism costs a lot of time in sending anti messages and state saving. This thesis presents an efficient optimistic protocol to minimize the rollbacks in the simulation system. We optimize the timewrap protocol in a way that the processes which are communicating and sending messages more rapidly bring closer to each other to improve the efficiency and to minimize the number of roll backs. To analyze the performance, the PHOLD application is used which is the standard PDES benchmark. The PHOLD provides the easily understandable controls to analyze the performance and test the functionality simultaneously.