Distributed Simulation of Large Systems

大型系统的模拟需要大量的计算和内存。然而，对于许多关键应用，仿真是设计高性能和可靠系统的唯一途径。分布式处理器和多处理器的广泛可用性使得通过在不同的处理器上分配模拟中的每个进程来执行模拟成为可能，因此，有可能实现显着的速度和每个处理器的低内存需求。然而，分布式仿真需要解决几个问题，包括分散时钟的进度、统计数据的收集、仿真的终止、与顺序仿真相比的开销分析以及性能优化。与以前使用集中式协调或使用分布式解决方案导致过多计算开销或过多通信开销的方法不同，在分布式仿真中开发了分层分散的问题解决方案。本项目的目标是探索直接从规范生成分布式仿真的方法，开发提高分布式仿真效率的技术，设计从分布式仿真中收集和分析数据的方法，并在各种多处理系统上进行实验。

Simulations of large systems are computation and memory intensive. However for many critical applications, simulation is the only way to design high performance and reliable systems. The wide availability of distributed processors and multiprocessors opens up the possibility of carrying out simulations by allocating each process in the simulation on a different processor, thus, potentially achieving significant speedups and low memory requirement per processor. However, distributed simulation requires resolution of several problems including decentralizing the advancement of clock, collection of statistics, termination of simulation, analysis of overheads as compared with sequential simulation, and optimization of performance. Unlike previous approaches that incur either too much computation overhead by using centralized coordination or too much communication overhead by using a distributed solution, hierarchical decentralized solutions to problems have been developed in distributed simulation. The objectives of this project are to explore methods for generating distributed simulations directly from specification, develop techniques for improving the efficiency of distributed simulations, design methods for collecting and analyzing data from distributed simulations, and carry out experiments on a variety of multiprocessing systems.