Escalonamento de tarefas imprecisas em ambiente distribuído

Real-time computing systems are defined as those systems subjected to timing constraints. In those systems, results must be not only logically correct but they also must be generated at the right moment. A basic problem one finds when building a distributed real time system is the allocation and scheduling of tasks on the available computing resources. There is an intrinsic difficulty in simultaneously achieving two fundamental goals: to guarantee that results are generated by the desired time and to make the system flexible enough so it can adapt to a dynamic environment and, that way, increase its own utility. The Imprecise Computation technique has been proposed in the literature as an approach to the scheduling of real-time systems. When this technique is used, each task has a mandatory part and an optional part. The mandatory part is able to generate a minimal quality result that is barely good enough to keep the system in a safe operational mode. The optional part refines the result until it achieves the desired quality level. This technique tries to conciliate the two fundamental goals mentioned above. Meanwhile, there is not in the literature a broad study on "how to solve the scheduling problem when real-time distributed systems are built based on Imprecise Computation concepts." The overall goal of this theses is to show how real-time applications, that are built upon Imprecise Computation concepts, can be scheduled in a distributed environment. We intend to show that Imprecise Computation concepts can be adapted to an environment where tasks execute in different processors and communication among them is done by sending messages. It is shown in the text that we can split this problem in the following four specific problems: • How to guarantee that mandatory parts will be finished before or at the respective task deadline, when we consider that tasks can execute in different processors and the use of messages creates precedence relations among them. • How to know that the execution of an optional part will not jeopardize the execution of previously guaranteed mandatory parts. • How to chose which optional parts should be executed when the resource "available processor time" is not enough to execute all of them. • How to decide which task runs on which processor, in a way that all mandatory parts can be guaranteed and that optional parts are evenly spread over the system so as to maximize the chance they get actually executed. This theses presents scheduling solutions for those four specific problems. In this way, the text shows that Imprecise Computation can effectively be used as the conceptual base for the construction of distributed real-time applications. Tese de Doutorado Rômulo Silva de Oliveira 1997