Real-Time Scheduling Algorithms

Real-time systems play an important role in our modern society. They are used in critical control systems that rely on timely response and computed results to function properly. This paper summarizes to an extent the state of real-time systems in the area of scheduling. A vast amount of work that has been done in the area of real-time scheduling by both the operations research and the computer science communities. While concerning ourselves with the scheduling requirements and methodologies of a real-time application, we observe that it is fundamentally different from a non time-constrained time-sharing application. The difference arises in the methods used for scheduling the tasks and the system support that is necessary for such an implementation. In this paper, we will consider the various issues that typify real-time scheduling. We discuss two major algorithms that forms a baseline for all scheduling approaches and we present a real-time implementation of such a system. Real-time systems differ from non-real-time systems in that they react to events of the physical world within a certain duration of time. Such a real-time system monitors and controls external processes, and must react to changes in a timely fashion, sometimes in the order of milliseconds. For simple enough processes the response is quick enough, but for many real-time systems that are more complex, a sophisticated coordination is required and rapid response to events becomes challenging. This motivates the study of methods to schedule various real-time events[1] so that time critical events are processed and responses are available to these real-time systems. These responses are critical because failure to respond to such events might result in failure of the system leading to loss of life and property. Two important properties of a task are the nature of the deadline and the arrival characteristics. The task arrival may be periodic with a constant interval, or the task may be stastically distributed and the inter-arrival time may be random in nature. The important requirement is that a time constrained task should be completed before the deadline set for that task expires. Depending on the real-time system, the nature of the deadline maybe different. Tasks with hard deadlines must be completed before the specified time. If the deadline is missed, the utility of execution for such a task in nil. A hard deadline[1] that is missed results in a failure of the system, depending on the task. On the other hand, tasks that gradually degrade with the deadline are called soft deadlines[16]. Such tasks impose a more flexible restriction on the scheduler and missing deadlines will not cause a catastrophic failure[1]. In this paper, we are concerned with abstract timing requirements in a hard-real-time environment, and the different ways of mapping these requirements to tasks with timing constraints. The methods for scheduling based on the task, and the mechanisms for scheduling such tasks in a real-time kernel are discussed. In Section 2 we discuss the classic scheduling algorithms in a multiprogramming system. Section 3 gives us additional information about the nature of real-time scheduling and some of the problems associated with priority inversions, Section 4 delineates some of the solutions to resource reclaiming which is shown to improve the utilization of the CPU. Section 5 introduces us to the concept of designing compilers for realtime systems with some introduction to the task ordering that is carried out in such applications. In Section 6 we discuss the implementation of a real-time system; Mach. We finally conclude with discussion in Section 7 and Section 8. 2. Scheduling Algorithms: Early work was carried out by Liu and Layland[2] who presented scheduling algorithms for fixed and dynamic tasks. The rate monotonic algorithm was shown to be useful for fixed priority tasks, and the earliest-deadline-first and minimum laxity first algorithms was proved to be useful for dynamically changing tasks. This gives us a broad classification of scheduling algorithms namely: static priority [2] and dynamic priority scheduling[2]. In all the related work that followed, the above algorithms were the basis for further optimization. 2.1 Rate Monotonic Algorithm (RM) This is a fixed priority algorithm[2] and follows the philosophy that higher priority is given to tasks with the higher frequencies. Likewise, the lower priority is assigned to tasks with the lower frequencies. The scheduler at any time always choose the highest priority task for execution. By approximating to a reliable degree the execution times and the time that it takes for system handling functions, the behavior of the system can be determined apriori. The rate monotonic algorithm[2, 7, 9] can successfully schedule tasks in a static priority environment but it has bound of less that 100% efficiency. The CPU utilization of tasks τi where 1 ≤ i ≤ n, is computed as the ratio of worst case computing time Ci to the time period Ti. The total utilization of the CPU is computed as follows[2]: