A Statistical Characterization Approach to Estimate Software Execution Time in Multiprocessor Systems

As multiprocessor systems become increasingly popular, the timing performance must be carefully evaluated to enable sound design decisions before final implementation. Nonetheless, shared resources can cause timing behavior of software hardly predictable. This paper introduces a measurement based on performance capture method coupled with statistical inference to infer the detailed statistical characteristics. It supports simulating various work scene and companion. A set of experiments are presented to validate the introduced methods and different possible results are shown. Based on the inference result, the paper provides several solutions to representing the execution time in multi-core system.

[1]  Axel Legay,et al.  Building faithful high-level models and performance evaluation of manycore embedded systems , 2014, 2014 Twelfth ACM/IEEE Conference on Formal Methods and Models for Codesign (MEMOCODE).

[2]  Liliana Cucu-Grosjean,et al.  PROXIMA: Improving Measurement-Based Timing Analysis through Randomisation and Probabilistic Analysis , 2016, 2016 Euromicro Conference on Digital System Design (DSD).

[3]  Donald E. Thomas,et al.  Stochastic Contention Level Simulation for Single-Chip Heterogeneous Multiprocessors , 2010, IEEE Transactions on Computers.

[4]  G. Box,et al.  On a measure of lack of fit in time series models , 1978 .

[5]  Andy D. Pimentel,et al.  Calibration of Abstract Performance Models for System-Level Design Space Exploration , 2006, 2006 International Conference on Embedded Computer Systems: Architectures, Modeling and Simulation.

[6]  Jan Reineke,et al.  Impact of resource sharing on performance and performance prediction , 2013, 2014 Design, Automation & Test in Europe Conference & Exhibition (DATE).

[7]  D. Vose.  Risk Analysis: A Quantitative Guide , 2000 .