The recent world record PHOLD performance result suggests optimistic parallel discrete event simulators (OPDES) should be able to deliver superior performance at extreme scale in many application domains. In fact, programming for OPDES today is extremely hard because of the necessity to write reversing and commit methods, in addition to the normal forward method of a conservative implementation. A second issue in extreme scale simulation is dealing with dynamic load imbalance. In this paper we will describe our approach to addressing these issues, using: source-to-source compiler tools to create optimistic forward, reverse and commit methods solely from the conservative method implementation; the ROSS OPDES simulator; and the Charm++ run time platform for dynamic load balancing.
展开▼
机译:最近的世界纪录PHOLD性能结果表明,乐观的并行离散事件模拟器(OPDES)应该能够在许多应用领域中以极端规模提供卓越的性能。实际上,由于除了保守实现的常规正向方法之外,还必须编写反向和提交方法,因此当今的OPDES编程非常困难。极端规模仿真中的第二个问题是处理动态负载不平衡。在本文中,我们将使用以下方法描述解决这些问题的方法:源到源编译器工具仅从保守方法的实现中创建乐观的正向,反向和提交方法; ROSS OPDES模拟器;以及用于动态负载平衡的Charm ++运行时平台。
展开▼