Guaranteed dynamic priority assignment scheme for streams with (m, k)-firm deadlines

Abstract

A guaranteed dynamic priority assignment scheme for multiple realtime streams with (m, k)-firm deadlines is presented. Analytical and experimental studies show that the proposed scheme provides assurance of timeliness performance and relatively high quality of service compared to existing schemes.Introduction: A weakly-hard real-time system can afford to miss some deadlines during any time window, i.e. the occasional loss of some deadlines is usually acceptable [2 -4]. One example is real-time video stream transmission, where a source generates a stream of video frames, which are transmitted and played back at the destination. Each frame has its own deadline by which it must arrive at the destination. In this system, a few occasional missed deadlines do not cause significant degradation in video quality, provided that there are only a limited number of consecutive deadline misses. To precisely specify the weakly-hard real-time requirement, Hamdaoui and Ramanathan have defined an (m, k)-firm deadline as when the quality of service is tolerable, provided at least m frames in any window of k consecutive frames meet their deadlines [1]. A stream that violates its own (m, k)-firm deadline, i.e. there are fewer than m occurrences of deadline satisfaction in a window of k consecutive frames, introduces a dynamic failure. Thus, the probability of a dynamic failure is used to measure how often the stream provides lower quality of service than is required.For dealing with the problem of scheduling multiple real-time streams constrained by (m, k)-firm deadlines, Hamdaoui and Ramanathan proposed a dynamic priority assignment scheme, the distance-based priority scheme (DBP), that assigns priority based on the recent history of streams' dynamic failure occurrences. More specifically, DBP assigns a priority according to the minimum number of consecutive deadlinemisses that are required for the stream to fall into the dynamic failure state. A higher priority is given to a stream with a shorter distance to its dynamic failure state. However in [2] it was pointed out that DBP, which is a best-effort online scheduling algorithm, has two major restrictions: 1. it provides non-guaranteed timeliness performance, and 2. it only considers homogeneous stream sets with the same execution and inter-arrival times, etc. In addition, the objective of stream scheduling is not only to provide guaranteed performance by avoiding dynamic failures of streams constrained by (m,k)-firm deadlines, but also to provide the highest possible quality of service, i.e. as many occurrences of deadline satisfaction of the stream as possible. To address these issues, we propose the guaranteed dynamic priority assignment scheme (GDPA) that schedules multiple streams constrained by (m, k)-firm deadlines. GDPA is designed to: 1. provide guaranteed real-time performance for multiple streams with (m, k)-firm deadlines when the system is underloaded; 2. reduce the probability of dynamic failures, and 3. maximise the probability of deadline satisfactions.