A number of library-based parallel and sequential network simulators have been designed. This paper describes a library, called GloMoSim (for Global Mobile system Simulator), for parallel simulation of wireless networks. GloMoSim has been designed to be extensible and composable: the communication protocol stack for wireless networks is divided into a set of layers, each with its own API. Models of protocols at one layer interact with those at a lower (or higher) layer only via these APIs. The modular implementation enables consistent comparison of multiple protocols at a given layer. The parallel implementation of GloMoSim can be executed using a variety of conservative synchronization protocols, which include the null message and conditional event algorithms. This paper describes the GloMoSim library, addresses a number of issues relevant to its parallelization, and presents a set of experimental results on the IBM 9076 SP, a distributed memory multicomputer. These experiments use mo...

Maisie is a C-based discrete-event simulation language that was designed to cleanly separate a simulation model from the underlying algorithm (sequential or parallel) used for the execution of the model. With few modifications, a Maisie program may be executed using a sequential simulation algorithm, a parallel conservative algorithm or a parallel optimistic algorithm. The language constructs allow the runtime system to implement optimizations that reduce recomputation and state saving overheads for optimistic simulations and synchronization overheads for conservative implementations. This paper presents the Maisie simulation language, describes a set of optimizations and illustrates the use of the language in the design of efficient parallel simulations. 1 Introduction Distributed (or parallel) simulation refers to the execution of a simulation program on parallel computers. A number of algorithms[25, 10, 11, 21, 20] have been suggested for distributed simulation and many experimental...

e-j 4r.,,D I-- " h",' _ k,) r,m '3'-. IC,-.-4 Z _ O

This paper describes the design and implementation of MPI-SIM, a library for the execution driven parallel simulation of MPI programs. MPI-LITE, a portable library that supports multithreaded MPI is also described. MPI-SIM, which is built on top of MPI-LITE, can be used to predict the performance of existing MPI programs as a function of architectural characteristics including number of processors and message communication latencies. The simulation models can be executed sequentially or in parallel. Parallel executions of MPI-SIM models are synchronized using a set of asynchronous conservative protocols. MPI-SIM reduces synchronization overheads by exploiting the communication characteristics of the program that it simulates. The paper presents validation and performance results from the use of MPI-SIM to simulate applications from the NAS Parallel Benchmark suite. Using the techniques described in this paper, we were able to reduce the number of synchronizations in the parallel simula...

Most work to date in parallel and distributed discrete event simulation is based on assigning precise time stamps to events, and time stamp order event processing. An alternative approach is examined where modelers use time intervals rather than precise time stamps to specify uncertainty as to when events occur. Partial orderings called approximate time (AT) and approximate time causal (ATC) order are proposed and synchronization algorithms developed that exploit these specifications to yield more efficient execution on parallel and distributed computers. Performance measurements of the AT-ordering mechanism on a cluster of workstations demonstrate as much as twenty-fold performance improvement compared to time stamp ordering with negligible impact on the results computed by the simulation. The context for much of this work is federated simulation systems that provided the initial motivation for this work. These results demonstrate that exploiting temporal uncertainty inhere...

Parallel discrete event simulation offers significant speedup over the traditional sequential event list algorithm. A number of conservative and optimistic algorithms have been proposed and studied for parallel simulation. We examine the problem of transparent execution of a simulation model using conservative algorithms, and present experimental results on the performance of these transparent implementations. The conservative algorithms implemented and compared include the null message algorithm, the conditional-event algorithm, and a new algorithm which is a combination of these. We describe how dynamic topology can be supported by conservative algorithms. Language constructs to express lookahead are discussed. Finally, performance measurements on a variety of benchmarks are presented, along with a study of the relationship between model characteristics like lookahead, communication topology and the performance of conservative algorithms. 1 INTRODUCTION Distributed(or parallel) simu...

Synchronization is often the dominant cost in conservative parallel simulation, particularly in simulations of parallel computers, in whichlow-latency simulated communication requires frequent synchronization. This thesis presents local barriers and predictive barrier scheduling,two techniques for reducing synchronization overhead in the simulation of message-passing multicomputers. Local barriers use nearest-neighbor synchronization to reduce waiting time at synchronization points. Predictive barrier scheduling, a novel technique whichschedules synchronizations using both compile-time and runtime analysis, reduces the frequency of synchronization operations. These techniques were evaluated by comparing their performance to that of periodic global synchronization. Experiments show that local barriers improve performance by up to 24% for communication-bound applications, while predictive barrier scheduling improves performance by up to 65% for applications with long local computation phases. Because the two techniques are complementary, I advocate a combined approach. This work was done in the context of Parallel Proteus, a new parallel simulator of message-passing multicomputers.

We describe a new algorithm, called Filter, that limits the propagation of erroneous computations in optimistic discrete-event distributed simulations. In the proposed algorithm, each message carries a bounded amount of dependency information that describes the assumptions made in the generation of the message, and, in addition, processes keep track of straggler events that have occurred in the system. This knowledge is used by processes to \ lter " out messages that depend on a preempted state by discarding them upon receipt. We describe the algorithm and its use in conjunction with time-warp, suggest several ways of reducing its potential overhead by adjusting the extent of ltering, and point out several interesting performance tradeo s that we are currently exploring. 1

The performance of parallel discrete event simulation protocols is heavily dependent on the lookahead of the simulation model. Identifying and expressing correct lookahead for a model isn't easy, nor is it a well-defined process. In this paper, a global view of a PDES model as a set of data flows is presented. Using this view, we show how the lookahead of the model can be optimized, and we present various simplified implementations of this global view and the significant performance improvements generated when applied to real world models. 1 Introduction The complexity of many modern systems, ranging from wireless networking, internetworking, wargaming, to computer architecture design, defies analytical modeling techniques. A scalable simulation environment, with features such as object aggregation, varying levels of abstraction, and parallel model execution, is necessary to model these systems. Further, in some of these domains, there is a distinct need for real-time responsiveness...

. Accurate simulations of parallel programs for large datasets can often be slow; parallel execution has been shown to offer significant potential in reducing the execution time of many discrete-event simulators. In this paper, we describe the design and implementation of a parallel simulator called DPSIM that simulates the execution of data parallel programs on contemporary message-passing parallel architectures. The simulator has been implemented on the IBM SPx using a conservative synchronization algorithm. This paper also describes the use of the simulator in evaluating the impact of architectural characteristics like processor speed and message communication latency on the performance of scientific applications including Gauss Jordan elimination and matrix multiplication. 1 Introduction Simulators for parallel programs can be effectively utilized to test, debug, and predict performance of parallel programs on a diverse set of parallel architectures. A variety of simulators have b...