Adaptive Mechanisms for Parallel Simulation

本项目将研究自适应同步协议，该协议监视并行模拟器的执行，并自动调整模拟器的执行以优化性能。 自适应机制允许底层仿真系统实现最大性能，特别是对于实践中经常出现的动态变化的工作负载。 这些机制还可以简化并行仿真软件的开发，使得某种程度的性能“调整”由仿真系统自动执行，而不是由应用程序员自动执行，应用程序员可能不是并行仿真器性能方面的专家。 提出了一种使用内存分配来控制执行的协议。 将解决的第二个问题涉及分析和定义合适的增量状态保存方法，这些方法被设计为与基于回滚的并行和分布式模拟器一起有效地执行。 这两个问题都将使用在多处理器硬件上运行的实现实验和分析建模技术来解决。

This project will examine adaptive synchronization protocols that monitor the execution of the parallel simulator, and automatically adapt the simulator's execution to optimize performance. Adaptive mechanisms allow the underlying simulation system to achieve maximum performance, particularly for dynamically changing workloads that often arise in practice. These mechanisms can also simplify the development of parallel simulation software to the extent that some degree of performance "tuning" is automatically performed by the simulation system, rather than by the application programmer, who may not be an expert in parallel simulator performance. A protocol that uses memory allocation to control execution is proposed. A second problem that will be addressed concerns analyzing and defining suitable incremental state saving methods that are designed to execute efficiently with rollback-based parallel and distributed simulators. Both problems will be attacked using experimentation on implementations running on multiprocessor hardware, and analytic modeling techniques.