Applicability of Extreme Value Theory to the Execution Time Prediction of Programs on SoCs

This paper describes in detail the estimation algorithm of upper bound prediction of the time acquisition task. We use the specific hardware from ARM Cortex-A series and empirical approach of time values retrieval from the timer counter. The robust Measurement-Based Probabilistic Timing Analysis (MB...

Full description

Bibliographic Details
Main Authors: Irina Fedotova, Bernd Krause, Eduard Siemens
Format: Article
Language:English
Published: Anhalt University of Applied Sciences 2017-03-01
Series:Proceedings of the International Conference on Applied Innovations in IT
Subjects:
Online Access:https://icaiit.org/paper.php?paper=5th_ICAIIT/S3_2
Description
Summary:This paper describes in detail the estimation algorithm of upper bound prediction of the time acquisition task. We use the specific hardware from ARM Cortex-A series and empirical approach of time values retrieval from the timer counter. The robust Measurement-Based Probabilistic Timing Analysis (MBPTA) method based on the Extreme Value Theory (EVT) has been used for experimental verification of the algorithm. The MBPTA method allows deriving a reliable and safe worst-case execution time (WCET) estimation based on the limited number of measurements on the target platform. However, it requires an appropriate complete set of statistical tests for verifying EVT applicability. In ongoing work, we intend to outline challenges behind EVT assumptions and parameter tuning for timing analysis, and provide more coherent approach for safe probabilistic WCET estimations in order to increase the confidence that timing constraints will be met.
ISSN:2199-8876