The authors study the adaptation of an optimistic Time Warp kernel to cross-cluster computing on the Grid. Wide-area communication, the primary source of overhead, is offloaded onto dedicated routing processes. This a...
详细信息
The authors study the adaptation of an optimistic Time Warp kernel to cross-cluster computing on the Grid. Wide-area communication, the primary source of overhead, is offloaded onto dedicated routing processes. This allows the simulation processes to run at full speed and thus significantly decreases the performance gap caused by the wide-area distribution. Further improvements are obtained by employing message aggregation on the wide-area links and using a distributed global virtual time algorithm. The authors achieve many of their objectives for a cellular automaton simulation with lazy cancellation and moderate communication. High communication rates, especially with aggressive cancellation, present a challenge. This is confirmed by the experiments with synthetic loads. Even then, a satisfactory speedup can be achieved, provided that the computational grain of events is large enough.
Parallel Discrete Event Simulations (PDES) running at large scales involve the coordination of billions of very. ne grain events distributed across a large number of processes. At such large scales optimistic synchron...
详细信息
ISBN:
(纸本)9781450350921
Parallel Discrete Event Simulations (PDES) running at large scales involve the coordination of billions of very. ne grain events distributed across a large number of processes. At such large scales optimistic synchronization protocols, such as TimeWarp, allow for a high degree of parallelism between processes, but with the additional complexity of managing event rollback and cancellation. This can become especially problematic in models that exhibit imbalance resulting in low event efficiency, which increases the total amount of work required to run a simulation to completion. Managing this complexity becomes key to achieving a high degree of performance across a wide range of models. In this paper, we address this issue by analyzing the relationship between synchronization cost and event efficiency. We first look at how these two characteristics are coupled via the computation of Global Virtual Time (gvt). We then introduce dynamic load balancing, and show how, when combined with low overhead gvt computation, we can achieve higher e. ciency with less synchronization cost. In doing so, we achieve up to 2X better performance on a variety of benchmarks and models of practical importance.
暂无评论