Performance Model for a Conservative Distributed Simulation Environment Using Null Messages to Avoid Deadlock

Abstract

A conservative distributed simulation requires all logical processes (LPs) to follow the causality constraint requirement. This implies that all event-messages are processed in strictly timestamp order. Apart from the timestamp of each event generated by LPs, synchronization between all LPs is the second most important requirements. Finally, there must not be a deadlock in the distributed environment. A deadlock may occur when there is no events present in the queue of LP. In such case, to avoid deadlock, Chandy-Misra-Bryant presented an algorithm called Null Message Algorithm (NMA) [3]. These null messages are passed as an event-message to other LPs and it stored in one of queues of LPs. This null message indicates that till the time stamp of that null message, all other events in the queue which have lesser time stamp than null message’s time stamp are safe to process. It means that there won’t be any arrival of any events from that logical process until current simulation time is equal to the time stamp of the null message. With the time stamp of the null message, a Lookahead value is added to the time stamp of that null message. This Lookahead value can be measure on certain kind of parameters such as delay to transmit a message, propagation delay, etc. therefore, calculating value of Lookahead is the most important part as Lookahead value affects the performance of the conservative distributed event simulation. Proper value of Lookahead can reduce the number of null messages which decreases the traffic of the network. In this paper, we demonstrate some calculation on the Lookahead which shows the performance of the distributed event simulation.