CRAW/P: a workload partition method for the efficient parallel simulation of manycores

  • Authors:

  • Shuai Jiao;Paolo Ienne;Xiaochun Ye;Da Wang;Dongrui Fan;Ninghui Sun

  • Affiliations:

  • SKL Computer Architecture, ICT, CAS, Beijing, P.R. China, Graduate University of Chinese Academy of Sciences, Beijing, P.R. China;École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland;SKL Computer Architecture, ICT, CAS, Beijing, P.R. China;SKL Computer Architecture, ICT, CAS, Beijing, P.R. China;SKL Computer Architecture, ICT, CAS, Beijing, P.R. China;SKL Computer Architecture, ICT, CAS, Beijing, P.R. China

  • Venue:
  • Euro-Par'12 Proceedings of the 18th international conference on Parallel Processing
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper addresses the workload partition strategies in the simulation of manycore architectures. The key observation behind this paper is that, compared to traditional multicores, manycores feature more non-uniform memory access and unpredictable network traffic; these features degrades simulation speed and accuracy of Parallel Discrete Event Simulators (PDES) when one uses static workload partition schemes. Based on the observation, we propose an adaptive workload partition method: Core/Router-Adaptive Workload Partition (CRAW/P). The method delivers more speedup and accuracy than static partition schemes by partitioning the simulation of on-chip-network independently from that of the cores and by synchronizing them differently. Using a PDES simulator, we evaluate the performance of CRAW/P in simulating a 256-core general purpose many-core processor. Running SPLASH2 benchmark applications, the experimental results demonstrate it can deliver speed improvement by 28%˜67% over static partition scheme and reduces timing errors to