Task Scheduling Algorithms For Energy-Efficient Multimedia Stream Computing

Multimedia stream computing, featuring such as video and audio processing, isbecoming the fastest growing workloads of embedded computer systems. This kind ofcomputing is not only computationally intensive with huge data access, but alsodelay-sensitive. So it poses a grand challenge for embedded multimedia streamingcomputing systems to guarantee the real-time requirements as well as to save more energy,especially for the battery-powered mobile devices. This paper investigates energy-efficientmulti-task scheduling algorithms for the video decoding, which is a typical streamcomputing application. It is of great significance for developing energy-efficient computingsystems.Based on analyzing the features of computing demands of video decoding tasks, thispaper proposes a dynamic voltage and frequency scaling (DVFS) based task schedulingalgorithm, which aims at reducing systems’ power consumption. It elucidates inherentmechanism leading to changes of computing demands by analyzing the process of videodecoding. And then, it analyzes the characteristics of this kind of change and its dynamicrange by experiments. It can be found that the energy-efficient task scheduling algorithmshould solve two key issues: to choose which frequency the processor should run at and todecide the priority of tasks, in order to save energy by using the so-called DVFSmechanism. According to the relationship of a convex function between frequency andpower, the principles to reduce power consumption are pointed out, that is, to reduce themagnitude and number of times of the changes in frequency. Based on accurate predictionof the computational demands, this paper proposes an algorithm: maximum of successiveaverage (MSA), to select processors’ running frequency, and uses the earliest deadline first(EDF) based priority algorithm to schedule priority level of tasks. The maximum canguarantee real-time feature and the successive average can enhance the stability and reducethe change times and amplitudes for the CPU running frequency. Meanwhile, it achieves adynamic scheduling mechanism by the EDF based priority algorithm, which makes thesystem process the tasks first with the earliest deadline. By doing so, it can improve thequality of service (QoS).Experimental results show that the proposed scheme combining the frequencyselecting algorithm of MSA and the EDF based priority algorithm has superiority inperformance, especially for the conditions when the task arrival time interval has relativelylarge variation. When the task arrival time interval changes with the network condition, itcan achieve optimal or close to optimal performance. Moreover, it performs very well withstability. The range of variation for system power consumption and the deadline miss-rate isblow1%and3%with the task arrival time intervals varying. Compared to the GRACE-OSscheme, the proposed solution performs much better both in the power consumption and thedeadline miss-rate. It reduces the power consumption by6%on average with a distinct reduction of the deadline miss-rate. Especially, when the task arrival time interval changeslargely, the deadline miss-rate of the solution is only7.3%to26.32%of GRACE-OS’s.When the GRACE-OS can not meet the real-time requirements of the multimedia tasks, theproposed solution still can achieve a rather good QoS.