Performance Analysis of Dynamic Priority Scheduling Algorithms in Real Time Systems

A scheduling algorithm is the one which decides a schedule for a given set of tasks. There are a number of algorithms for task scheduling on a processor. Some of these algorithms are used for scheduling tasks on a multiprocessor system either under the partitioning scheme or under the global scheme. The optimal algorithm is the Earliest Deadline First (EDF) algorithm. If a set of task cannot be scheduled under EDF, no other scheduling algorithm can feasible schedule this task. Many improved versions of the existing conventional EDF algorithm have been proposed until now. In this paper, a study on the conventional EDF is done in detail. Finally, an evaluation of EDF algorithm using the processor values and static values is done. Introduction Real time systems are the ones in which the correctness of a system depends not only on the results of computations but also on the time at which the system produces its results [1]. In real time systems, there are fixed time constraints for each task to be performed. If the tasks are not completed in the specified time, the system is considered to be failed. Taking this into context, we can broadly classify real time systems into two: Soft Real Time Systems and Hard Real Time Systems [1]. Real time systems in which nothing catastrophic will happen if the deadlines are not met by each task are called Soft Real Time Systems [2]. This will only affect the performance of the system. For e.g. Multimedia. Whereas hard real time systems are those in which the result might be disastrous if deadlines are not me[4]t. In order to make sure the deadlines are met, a task scheduler is used. Real time scheduling algorithms or techniques can be broadly categorized into two: dynamic and static. The priorities for tasks in a static algorithm are given at the design time. In a dynamic algorithm the priorities are assigned during the run time. The dynamic scheduling can again be subdivided into two categories, static priority and dynamic priority. The two prominent scheduling algorithms that fall into these two categories are Earliest Deadline First (EDF) and Rate Monotonic (RM) respectively [2]. The schedule may be preemptive, if a task can be interrupted by another task of higher priority or nonpreemptive, where the task must be run to completion until it gets interrupted. Both EDF and RM algorithms are preemptive schedules. Earliest deadline first EDF is an optimal uniprocessor algorithm. Here tasks are preemptible and the priority is based on the deadlines. The task with the earliest deadline will have the highest priority. If a set of tasks cannot be scheduled on a single processor by EDF algorithm, then there is no other algorithm that can successfully schedule these tasks[3]. Rate monotonic This algorithm can be overrated above all other algorithms used till today. It is a static priority algorithm which is used in uniprocessor systems. The task which has the shortest period will have the highest priority. Once the priority levels are fixed for each task, it remains the same until the end of scheduling. Bin packing algorithm This algorithm deals with task assignment of multiprocessor systems. It is based on the condition that the total processor utilization of each processor should not exceed a threshold [1,4]. The threshold value depends on the uniprocessor scheduling scheme used. In this paper, we use bin packing for EDF algorithm. Hence, in this case, uniprocessor utilization should not exceed 1. Problem statement This work is an analysis of Earliest Deadline First algorithm on a uniprocessor platform and extends the work to a multiprocessor. EDF is an optimal algorithm but when the scheduling load on the system is very high its performance decreases [3,6]. As mentioned earlier, when all tasks are independent the task that has the earliest deadline is taken at each scheduling point. Unlike the rate monotonic algorithm, this reason makes scheduling using EDF a tedious task. Even today, though EDF is an optimal algorithm, static priority algorithms are used in real time systems [5]. Certain terminology that must be understood before we get into the analysis is discussed below. For each event that occurs, it triggers a task which is to executed for that particular event [2]. Hence depending on the occurrence of the event the task will also [Martin et al., 2(1): January, 2015] ISSN: 23495197 INTERNATIONAL JOURNAL OF RESEARCH SCIENCE & MANAGEMENT http: // www.ijrsm.com (C) International Journal of Research Science & Management [2] be repeated. This can be either periodic or aperiodic. Each time a task occurs, it is called the instance of task. The time interval between time 0 and the instant where the actual deadline occurs is called absolute deadline, whereas Relative deadline is the time between the start of the task and the actual deadline [1]. Assumptions In the examples given further, while scheduling tasks following assumptions are made: 1. There are no non-preemptible sections in any task taken and the cost of the pre-emption is insignificant. 2. Only processing necessities are noteworthy; memory, I/O and other source requirements are insignificant. 3. There are no precedence constraints; all tasks are independent. 4. The period of a task is equal to its relative deadline. 5. All tasks taken are released at the same time in the beginning