63. Complexity and Accuracy Tradeoff Analysis of Parallel Application Simulation Using SST/Macro
Authors: Zhou Tong (Florida State University)
Abstract: Modeling and simulating of HPC applications on contemporary supercomputing platforms is challenging due to the growing size of systems and applications. To understand the performance characteristics, parallel applications are often simulated at different levels including flow-level, packet-level, and flit-level. It is commonly believed that more detailed simulation will result in more useful information about the applications. However, it remains unclear how much more accurate simulation results finer-grained simulations can deliver at the cost of higher simulation complexity. In this study, we investigate the trade-off between modeling and simulation complexity and accuracy using the macroscale component of the Structured Simulation Toolkit (SST) and a modeling tool. We measure the performance of a set of parallel applications on three current generation supercomputers, analyze the modeling and simulation results produced by the tools, and draw conclusions from the analysis of the statistics about the trade-off between complexity and accuracy of modeling and simulation.
Two-page extended abstract: pdf