Assessment of Trace-Differences in Timing Analysis for Complex Real-Time Embedded Systems

In this paper, we look at identifying temporal differences between different versions of Complex Real-Time Embedded Systems (CRTES) by using timing traces representing response times and execution times of tasks. In particular, we are interested in being able to reason about whether a particular change to CRTES will impact on their temporal performance, which is difficult to answer due to the complicated timing behavior such CRTES have. To be specific, we first propose a sampling mechanism to eliminate dependencies existing in tasks' response time and execution time data in the traces taken from CRTES, which makes any statistical inference in probability theory and statistics realistic. Next, we use a mature statistical method, i.e., the nonparametric two-sample Kolmogorov-Smirnov test, to assess the possible temporal differences between different versions of CRTES by using timing traces. Moreover, we introduce a method of reducing the number of samples used in the analysis, while keeping the accuracy of analysis results. This is not trivial, as collecting a large amount of samples in terms of executing real systems is often costly. Our evaluation using simulation models describing an industrial robotic control system with complicated tasks' timing behavior, indicates that the proposed method can successfully identify temporal differences between different versions of CRTES, if there is any. Furthermore, our proposed method outperforms the other statistical methods, e.g., bootstrap and permutation tests, that are often widely used in contexts, in terms of bearing on the accuracy of results when other methods have failed.