Reliable Energy-Aware Scheduling Algorithm With Multi-Level Budget for Real-Time Embedded System

Introduction

Real-time embedded systems are designed for mobile applications where minimization of response time and energy are essential. A co-optimized scheduling strategy is required to assure real-time efficient performance while making the most practical use of available energy resources. The current work deals with minimization of energy for time-critical applications that are dependent on battery and have requirements of both fixed arrival pattern and random arrival pattern. The system can operate at different speed levels. Minimization of energy issues is taken care of along with improvement in response time for random tasks. It also facilitates acceptance of more number of functions. Growth in feasibility analysis is applicable both for fixed as well as random pattern tasks. While looking to improve response time for aperiodic tasks as well as feasibility analysis for periodic tasks, designers force the system to operate at higher speed and reduce its requirement in terms of time. However, the requirement of reduction in time is achieved at the cost of enhanced power consumption that is, more energy requirement. On the other side, while reducing energy consumption, it forces the system to operate at a lower speed level, consuming less power as well as energy. This reduction in voltage at the cost of increase in time requirement of a task causes either inability to complete the job within the deadline, or an elongated response time in case of aperiodic jobs. Thus, we can see that energy reduction and improvement in response time are orthogonal issues. These issues can be tackled through trade-offs. The feasibility of periodic tasks and server is determined offline using Rate-monotonic first static priority assignment technique. The rate-monotonic algorithm assigns priorities based on the principle, lower the period, the higher its priority (Duan H. et al., 2017). For the case tasks that have same period, any task can be given top priority. Aperiodic tasks execute at server priority (assigned offline) and are limited only by availability of budget in terms of energy.

This research aims to reduce consumption of energy in an energy constrained real-time system, with a mixed task set, by utilizing the concept of dynamic voltage scaling and active power down. Decision for energy consumption (periodic task) is taken at both task as well as job levels. In contrast, runtime tuning is received on the level of availability of aperiodic requirements, as stated by Fan M. et al. (2018). The three-phase server is proposed to take care of dynamic as well as latent demand for time-critical applications, depending on the battery. Speed for phase I and phase II is fixed well in advance (offline), whereas dynamic speed tuning is carried out in phase III. Phase I ensures feasibility of periodic tasks through fixation of speed at the task level, whereas job level speed refinement is carried out in phase II. Further, in phase III, a tune speed accumulation of empty slots and priorities for execution of the budget are used to execute dynamic requirements. In this work, periodic and static conditions are used interchangeably, and dynamic requirements and aperiodic tasks are used as synonyms.

Significant contributions to this paper:

The main contribution of this paper is to develop a three-phase algorithm aimed at energy minimization for battery-operated applications. In the first phase, task level speed assignment is done with feasibility of each periodic task in the task set, to ensure response along with the server, to aperiodic tasks, and computing the budget. The second phase adopts job level speed assignment. Thus, speed tuning is proposed at the job level to reduce energy consumption. Again, energy consumption is reduced in the third phase by adjusting the speed assigned to a job and gathering of idle slots. This proposed three-phase approach provides better tolerance (approximately four times) and still consumes less energy (about 35% to 45% less) in a wide range of applications.

The rest of the paper is organized as follows; Section II provides a system energy model; the proposed three-phase server to take care of energy-constrained real-time application is given in Section III; simulation results and analysis are detailed in Section IV; finally, the paper concludes in Section V.