Utility accrual real-time scheduling for (m,k)-firm deadline-constrained streams on multiprocessors

Abstract

Considering video conferencing applications, presented is the first utility accrual (or UA) real-time scheduling algorithm for multiple (m, k)-firm deadline-constrained streams running on multiprocessors, called the global multiprocessor utility accrual scheduling algorithm for (m, k)-firm deadline-constraint multimedia streams (or gMUA-MK). Analytical and experimental studies show that gMUA-MK achieves timeliness performance and relatively high quality of multimedia services compared to existing schemes including gMUA.Introduction: Video conferencing is one of highly dynamic real-time multimedia services, where multiple sources generate multiple streams of video frames, which are transmitted and played back at the intended destinations. Video conferencing allows multiple users to dynamically join and leave the conference, which sometimes imposes excessive computational workloads on the destinations. In its operation, one participant is speaking while all the others are listening for a period of time, which implies that the streams containing the gestures and voice of the speaker are more important than others containing those of listeners. A video stream includes a series of frames constrained by their own deadlines. Intrinsically, a few occasional deadline misses of frames are tolerable without significant degradation of multimedia service quality when it is played back at the destinations. Similar features of highly dynamic real-time multimedia services are also found in remote medical imaging, video surveillance, etc.These features lead us to designing real-time scheduling algorithms considering that 1. an overloaded computational workload should be addressed with graceful performance degradation, 2. the difference in the importance of each stream should be distinguishably considered, and 3. the tolerance to a few occasional deadline misses should be precisely expressed and handled.The urgency of a stream, represented as the deadline, is typically orthogonal to its importance, e.g. the most urgent activity can be the least important, and vice versa. Since meeting the deadlines of all streams is impossible in overloads, completing the most important streams irrespective of stream urgency is often desirable. Thus, a clear distinction has to be made between urgency and importance during overloads. For this reason, we consider the abstraction of time/utility functions (TUFs) that express the utility of completing an application activity as a function of that activity's completion time. Particularly, we consider step TUFs where the height and the length of a TUF represent the importance and deadline of a task, respectively. When the time constraints are specified by TUFs, the scheduling criteria are based on accrued utility, such as maximising the sum of the activities' attained utilities, which is called the utility accrual (UA) criteria. The latest UA scheduling for multiprocessors is the global multiprocessor utility accrual scheduling (gMUA) in [1].Weakly-hard real-time systems are defined as the ones that can...