Evolutionary Techniques for Parametric WCET Analysis

Estimating the worst-case execution time (WCET) of real-time programs is pivotal in their verification. WCET estimation either yields a numeric value that represents the maximum execution time of the program when executed on a specific hardware platform; or yields a parametric expression in the form of some function of the input which when instantiated with a particular input value, gives a WCET estimation of the program when triggered by this input specifically (on a specific hardware platform). Parametric WCET analysis provides extra accuracy as the WCET estimation can be tuned to particular input values at runtime; and is usually of interest to dynamic-scheduling schemes. In this paper we use genetic programming as an alternative method to approach the problem of parametric WCET analysis. Parametric expressions are captured automatically by the genetic program based on end-to-end executions of the program under analysis. The technique is complementary to static parametric WCET analysis and more amenable to industrial practice. Experimental evaluation shows that the presented technique computes accurate parametric expression in an almost negligible time