Predicting Worst-Case Execution Time Trends in Long-Lived Real-Time Systems
In some long-lived real-time systems, it is not uncommon to see that the execution times of some tasks may exhibit trends. For hard and firm real-time systems, it is important to ensure these trends will not jeopardize the system. In this paper, we first introduce the notion of dynamic worst-case execution time (dWCET), which forms a new perspective that could help a system to predict potential timing failures and optimize resource allocations. We then have a comprehensive review of trend prediction methods. In the evaluation, we make a comparative study of dWCET trend prediction. Four prediction methods, combined with three data selection processes, are applied in an evaluation framework. The result shows the importance of applying data preprocessing and suggests that non-parametric estimators perform better than parametric methods.
KeywordsWorst-case execution time Trend prediction Linear regression Extreme value theory Support vector regression
- 2.Fisher, R.A., Tippett, L.H.C.: Limiting forms of the frequency distribution of the largest or smallest member of a sample. In: Mathematical Proceedings of the Cambridge Philosophical Society, vol. 24, pp. 180–190. Cambridge Univerisity Press (1928)Google Scholar
- 10.Reinsel, G.C., Tiao, G.C.: Impact of chlorofluoromethanes on stratospheric ozone: a statistical analysis of ozone data for trends. J. Am. Statist. Assoc. 82(397), 20–30 (1987)Google Scholar
- 15.Stankovic, J.A., Lu, C., Son, S.H., Tao, G.: The case for feedback control real-time scheduling. In: Proceedings of the 11th Euromicro Conference on Real-Time Systems, pp. 11–20. IEEE (1999)Google Scholar
- 18.Wilhelm, R., Engblom, J., Ermedahl, A., et al.: The worst-case execution-time problem overview of methods and survey of tools. ACM Trans. Embedded Comput. Syst. (TECS) 7(3), 36 (2008)Google Scholar