Development of Feedback Controlled System Architecture for Scheduling Hard and Soft Real Time Tasks on a Multiprocessor Platform

Abstract

This thesis focuses on using feedback to tackle uncertainty in real time scheduling on multiprocessor platform. A system can be heavily underutilized when used to schedule unpredictable load. We make use of feedback to mitigate the effect of such unpredictable load and utilize the resources effectively. Since most of the systems have more than one type of real time tasks, we consider scheduling hard and soft real time tasks together. It has been shown that the system performance can be considerably improved when feedback is used with scheduling algorithms. Previous work on feedback scheduling is focused on a single type of tasks in the system. The proposed algorithm is able to schedule two types of task and guarantees performance of both the applications independently. Moreover, with the proposed architecture any algorithm used for multiprocessor scheduling can be plugged-in in the system to make it close loop one. This makes our architecture generic. It has been proved that schedulability is guaranteed for hard real time tasks and tardiness is bounded for soft real time tasks with our system. The results have been verified by using global-EDF for scheduling the taskset. The system has been further evaluated with mixed criticality taskset. A variant of fpEDF algorithm is used alongwith feedback, to make it a closed loop algorithm. The results prove that better schedulability can be achieved with feedback controlled fpEDF than the conventional one. We argue on making all open loop algorithms as closed loop ones to handle unpredictable load.