Parallel Discrete-Event Simulation

Parallel discrete-event simulation (PDES), simply referred to as parallel simulation, is concerned with the execution of discrete-event simulation on parallel computers. PDES has been recognized as a challenging research field bridging between modeling and simulation, and high-performance computing. By exploiting the potential parallelism in a simulation model, PDES can overcome the limitations imposed by sequential simulation both in the execution time and the memory space, and therefore demonstrate as a viable technique for solving large-scale complex models. In this article, we provide a brief overview of the current state of PDES, identify its fundamental challenges, and discuss existing principal solutions resulted from three decades of intensive research in this field. Further, we report specific research advances in high-performance modeling and simulation of large-scale computer networks as an exemplar of typical PDES applications. 1 Simulation and Parallelization Methods 1.1 Simulation with Discrete Events Simulation—the practice of mimicking the operations of systems over time— is one of the most important and widely used techniques in operations research. These systems are abstracted as models in the form of mathematical or logical relationships. In simulation, one uses computers to evaluate the models numerically in a controlled environment, where data are gathered and used to estimate the behavior of the target systems. Simulation is effective for prototyping new system designs, as well as providing insight to the true characteristics of existing systems. Simulation is particularly indispensable for studying large-scale and complex systems, which can be otherwise intractable to closed-form mathematical or analytical solutions. Due to the practical nature of the design process where simplified assumptions fly in the face of required complexities, simulation is an irrefutable choice for both system design and evaluation. A simulation model specifies the state evolution over time. The target system can be viewed either as a continuous system, where state changes continuously with respect to time, as a discrete system, where state is modified only at specific points in time, or as a hybrid system, which consists of both continuous and discrete elements. In simulation, time progression is described by the so-called “time advancement function”, which specifies two classes of simulation methods: time-driven and event-driven. In a timedriven (or time-stepping) simulation, time is measured at small intervals giving the impression that the system evolves continuously over time. As such, it is naturally more suitable for simulating continuous systems. In an event-driven (or discrete-event) simulation, time leaps through distinct points in time, which we call events. Consequently, event-driven simulation is more appropriate for simulating discrete systems. Note that one can combine both types of simulations, for example, in a computer network simulation, using discrete events to represent detail network transactions, such as sending and receiving packets, and using continuous simulation to capture the fluid dynamics of overall network traffic [34]. This article mainly focuses on discrete-event simulations and efficient techniques to parallelize them. It is necessary to first understand what constitutes a sequential discrete-event simulation before we proceed to review the parallelization methods. A discrete-event simulation maintains a data structure called the event-list, which is basically a priority queue that sorts events according to the time at which they are scheduled to happen in the simulated future. A clock variable T is used to denote the current time in simulation. At the heart of the program is a loop; the simulator repeatedly removes an event with the smallest timestamp from the eventlist, sets the clock variable T to the timestamp of this event, and processes the event. Processing an event typically changes the state of the model and may generate more future events to be inserted into the event-list. The loop continues until the simulation termination condition is met, for example, when the event-list becomes empty or when the simulation clock has reached a designated simulation completion time.