In this article we develop an automatic deadlock detection scheme for general purpose simulation systems and offer deadlock-resolution algorithms for grouped resources and overlapping resources. Our intent is to provide capabilities that can be incorporated in professional simulation systems to facilitate modeling situations where deadlocking is a problem. Two general modeling situations are analyzed. In grouped-resources models, a set of resources is required to complete a task, whereas in overlapping-resources models, movement to the next process requires first gaining control of the next resource. Deadlocks occurring in these two different situations require different resolution algorithms. We develop a graph-theoretic description of the simulation over time and define deadlocks in terms of strongly connected components of this graph. We make a distinction between transient deadlocks and permanent deadlocks. A prototype implementation of the detection and resolution algorithms was incorporated in a simulation system to provide information concerning the computational requirements and performance of these algorithms. We demonstrate that resolving a possible transient deadlock before it becomes permanent is worthwhile in conjunction with overlapping resources, and that the computational performance of the approach is reasonable except for situations with frequent occurrences of overlapping-resources deadlocks.

Deadlock is a situation when a set of entities, holding some resource(s), all request access to resource(s) held by other entities in the same set. Due to the stochastic nature of simulation and a lack of information of the chronology of events a priori, predicting deadlocks in simulation is a difficult task. In this paper, we present a graph theoretic model to capture the deadlock characteristics, and some deadlock properties of a general discrete simulation system. Finally, we suggest deadlock detection and resolution as an appropriate strategy in simulation systems.>

A classical method of measunng the number of lobs m a trine-shared computer system ,s to collect statistics at the epochs of quantum completions.A computer system utilizing round-robin quantum allocations ~s modeled as a feedback queue with Poisson amvals and exponential service times, and the bias in the stausuc taken at the quantum compleuons is quant,fied.The difference botween this statistic and the true Ume-average system size is given for such a system including quanta and overhead.