Research On WCET Estimation Of Real-Time System Software Based On Extreme Value Statistic

We must emphasis the importance of the execution time of task in the embed-system, especially in the real-time system design. It is needed to confirm the system time-limit or not, namely, the task must be finished before the deadline for the system safely running, or the real-time system will be collapsed. The most important parameter judging the system is the Worst-Case Execution Time (WCET), which supplies warrants for task real-time scheduling, judging task PRI, resource collision, and corresponding between tasks as the creditable base that keeping the system running safely.The purpose of WCET estimation is getting the right limit of the certain task, which should consider the software and hardware configuration comprehensively. For the dynamic measure method of WCET can not pledge the safety of the estimation, most researchers apply themselves to analysis WCET by static method at present-analyzing all possible flow and hardware configuration, which influences the execution time of the program. However, the static analysis needs some extra knowledge and limited suppose on system to a certain extent, which will affect the precision of WCET estimation, which makes the current static method over-highly in substance.We introduce a method of statistic based on generalized extreme value distribution (GEV) and generalized Pareto distribution (GPD) in the estimation of WCET, and proposed a statistical model of WCET and the methods of parameter estimation and checkout for the model. To compare with former methods, the one we mentioned will show all hardware configurations influencing on WCET of the program and more precise, which is applicable to the WCET estimation of the program with complicated hardware and software configuration. The results of the experiment indicate that the WCET estimation models constructed by generalized extreme value distribution (GEV) and generalized Pareto distribution (GPD) can supply WCET estimation of the real-time program quickly and effectively. At last, the advantages and the disadvantages of the two statistic analysis methods are compared.