This thesis focuses on scheduling tasks in a heterogeneous environment with DVS enabled processors to minimize both execution time and energy consumed. The proposed algorithm, called Energy Dynamic Level Scheduling (EDLS), favors low-energy consuming processors by introducing a cost factor that affects scheduling decisions.
"In the past 20-some years, the entire lifetime of Data Center, the hymn computer engineers and end users have chanted in harmony has been "faster. . .smaller. . . cheaper. . . lower power. . . ," with the most recently added "and lower temperature. . ." significantly complicating the whole scenario. The trade offs among performance, complexity, cost, power and temperature have created exciting challenges and opportunities. All modern data centers face the widespread problem "High performance without trading energy, power and most important temperature". Previous research on scheduling algorithms of processors have focused on static implementation to minimize energy consumption and heat dissipation, but never used Machine Learning to dynamically apply the algorithm. We use Naive Bayesian Classifiers (NBCs) to select the processor combination for the Temperature and Energy Aware Dynamic Level Scheduling algorithm that satisfies a particular user defined condition such as a deadline, energy or temperature budget. Our simulation results exhibit significant energy and temperature savings at a reasonable increase in overall execution time, the learning algorithm selects the desired processors significantly faster than random selection." -- page iv.
Multiprocessor platforms play important roles in modern computing systems, and appear in various applications, ranging from energy-limited hand-held devices to large data centers. As the performance requirements increase, energy-consumption in these systems also increases significantly. Dynamic Voltage and Frequency Scaling (DVFS), which allows processors to dynamically adjust the supply voltage and the clock frequency to operate on different power/energy levels, is considered an effective way to achieve the goal of energy-saving. This book surveys existing works that have been on energy-aware task scheduling on DVFS multiprocessor platforms. Energy-aware scheduling problems are intrinsically optimization problems, the formulations of which greatly depend on the platform and task models under consideration. Thus, Energy-aware Scheduling on Multiprocessor Platforms covers current research on this topic and classifies existing works according to two key standards, namely, homogeneity/heterogeneity of multiprocessor platforms and the task types considered. Under this classification, other sub-issues are also included, such as, slack reclamation, fixed/dynamic priority scheduling, partition-based/global scheduling, and application-specific power consumption, etc.
"With widespread growth of the internet, data centers have been trying to incorporate faster and more powerful processors to meet the high data processing requirements. Faster and more powerful processors have resulted in higher heat dissipation and energy consumption, which have had detrimental effects on operating costs. High heat dissipation, however, also affects the reliability of processing units. Previous research on scheduling algorithms of processors have either focused on minimal energy consumption or minimal heat dissipation, but never a combination of both. Hence, the focus of this research has been to schedule tasks in a heterogeneous environment with Dynamic Voltage Scaling (DVS) enabled processors to minimize execution time, energy consumption and heat dissipation. Our proposed algorithm, Temperature and Energy aware Dynamic Level Scheduling (TEDLS), favors the cooler and more energy efficient processors by introducing a cost function that affects the scheduling decisions." -- leaf v.
Multi-core processors increasingly appear as an enabling platform for embedded systems, e.g., mobile phones, tablets, computerized numerical controls, etc. The parallel task model, where a task can execute on multiple cores simultaneously, can efficiently exploit the multi-core platform’s computational ability. Many computation-intensive systems (e.g., self-driving cars) that demand stringent timing requirements often evolve in the form of parallel tasks. Several real-time embedded system applications demand predictable timing behavior and satisfy other system constraints, such as energy consumption.
Real-time embedded systems have become ubiquitous in our daily life. Due to their diversified usage, the research on these systems has confronted with many emerging challenges. One such challenge is to reduce power and energy consumption while maintaining assurance that timing constraints will be met. Power densities in microprocessors are almost doubled every three years. As energy is power integrated over time, supplying the required energy may become prohibitively expensive, or even technologically infeasible. This is particularly difficult in portable systems that heavily rely on batteries for energy, and will become even more critical as battery capacities are increasing at a much slower rate than power consumption. This book presents four contributions that are based on the thesis that energy-efficiency of Real-time Embedded Systems and scheduling are closely related problems and therefore, should be tackled together for optimal results. Contributions of this book are: 1) Two-level Hierarchical Scheduling Algorithm for Multiprocessor Systems, 2) Assertive Dynamic Power Management Scheme, 3) Deterministic Stretch-to-Fit DVFS Technique, and 4) Hybrid Power Management Scheme.
