Ph.D. Thesis
Depuración del comportamiento temporal de sistemas de tiempo real posix basada en mediciones
Author: J. Entrialgo
Institution: University of Oviedo (Spain)
Year: 2006
Date: 2006-03-31
Supervised by: J. García
Abstract
Time behaviour of real-time systems is of special importance. This has lead to the development of techniques which aim to ensure that the time requirements are fulfilled. Planificability analysis methods, which use a model of the system to determine if it is possible to obtain results while meeting the assigned deadlines, stand out among these techniques. In addition, monitoring systems have been developed in order to measure the times of the implemented system. Planificability analysis methods and monitoring systems do not usually share information, which makes the debugging of time behaviour errors complicated.
In this thesis, a method to make the debugging of time behaviour errors easier is presented. This method relates the data from the models used in the planificability analysis to the measurements obtained from the system. The method is applied to two types of model: on one hand, a "traditional" model based on calculating the worst response time of the tasks of the system from their worst execution time; on the other hand, a probabilistic model, which calculates the probability of not fulfilling the deadlines of the tasks in the system from a probability distribution of their execution times. The method proposed in this thesis makes it easier to pinpoint behaviour time errors automatically, by comparing the values in the model with the measurements. This comparison can become complicated in systems analyzed with probabilistic models, as two probability distribution functions must be compared. In order to carry out this comparison, two heuristic metrics, "pessimism" and "optimism", have been proposed. They are the first approach to a complex problem which until now has not been tackled in the real-time system research.
In order to check the applicability of the method, a tool for debugging the time behaviour of POSIX real-time systems has been implemented. This tool has been used with a case study carried out on a multiprocessor cluster. The tool has also been extended to compare probability distributions visually.
Finally, several case studies are carried out, which make it possible to analyze both the advantages of the techniques presented in this thesis and how the metrics pessimism and optimism work.
