Statistical performance analysis with dynamic workload using S-NET

V. Wieser, P. Holzenspies, R. Kirner, M. Roßbory. Statistical performance analysis with dynamic workload using S-NET. 1, 2012.

  • Volkmar Wieser
  • Philipp K.F. Holzenspies
  • Raimund Kirner
  • Michael Roßbory
NoteWorkshoppaper at Workshop on Feedback-Directed Compiler Optimization for Multi-Core Architectures (FD-COMA 2012) at HiPEAK’12 Conference, Paris, France, January 23-25, 2012

In this paper the ADVANCE approach for engineering concurrent software systems with well-balanced hardware efficiency is addressed using the stream processing language S-NET. Therefore the load distribution of S-NET is analyzed, comparing the statical versus dynamical workload whereat the main interest is not in runtime performance itself but in load balancing. To obtain the cost information in the concurrent system the metrics throughput, latency, and jitter are evaluated by analyzing generated synthetical data as well as using an industrial related application in future. The presented results of the current work are on the one hand an indicator for the status quo of the ADVANCE vision and on the other hand used to improve the applied statistical analysis techniques within ADVANCE. Like the ADVANCE project, this work is still under development, but further improvements and speedups are expected in the near future.