Dynamic voltage and frequency scaling (DVFS) is an extensively studied energy manage ment technique, which aims to reduce the energy consumption of computing platforms by dynamically scaling the CPU frequency. Real-Time DVFS (RT-DVFS) is a branch of DVFS, which reduces CPU energy consumption through DVFS, while at the same time ensures that task time constraints are satisfied by constructing appropriate real-time task schedules. The literature presents numerous RT-DVFS scheduling algorithms, which employ different techniques to utilize the CPU idle time to scale the frequency. Many of these algorithms have been experimentally studied through simulations, but have not been implemented on real hardware platforms. Though simulation-based experimental studies can provide a first-order understanding, implementation-based studies can reveal actual timeliness and energy consumption behaviours. This is particularly important, when it is difficult to devise accurate simulation models of hardware, which is increasingly the case with modern systems. In this thesis, we study the timeliness and energy consumption behaviours of fourteen state- of-the-art RT-DVFS schedulers by implementing and evaluating them on two hardware platforms. The schedulers include CC-EDF, LA-EDF, REUA, DRA andd AGR1 among others, and the hardware platforms include ASUS laptop with the Intel I5 processor and a mother- board with the AMD Zacate processor. We implemented these schedulers in the ChronOS real-time Linux kernel and measured their actual timeliness and energy behaviours under a range of workloads including CPU-intensive, memory-intensive, mutual exclusion lock-intensive, and processor-underloaded and overloaded workloads. Our studies reveal that measuring the CPU power consumption as the cube of CPU frequency can lead to incorrect conclusions. In particular, it ignores the idle state CPU power consumption, which is orders of magnitude smaller than the active power consumption. Consequently, power savings obtained by exclusively optimizing active power consumption (i.e., RT-DVFS) may be offset by completing tasks sooner by running them at the highest frequency and transitioning to the idle state earlier (i.e., no DVFS). Thus, the active power consumption savings of the RT-DVFS techniques' that we report are orders of magnitude smaller than their simulation-based savings reported in the literature. / Master of Science

Several modern multi-core architectures support the dynamic control of the CPU's clock rate, allowing processor cores to temporarily operate at speeds exceeding the operational base frequency. Conversely, cores can operate at a lower speed or be disabled altogether to save power. Such facilities are notably provided by Intel's Turbo Boost and AMD's Turbo CORE technologies. Frequency control is typically driven by the operating system which requests changes to the performance state of the processor based on the current load of the system. In this paper, we investigate the use of dynamic frequency scaling from user space to speed up multi-threaded applications that must occasionally execute time-critical tasks or to solve problems that have heterogeneous computing requirements. We propose a general-purpose library that allows selective control of the frequency of the cores - subject to the limitations of the target architecture. We analyze the performance trade-offs and illustrate its benefits using several benchmarks and real-world workloads when temporarily boosting selected cores executing time-critical operations. While our study primarily focuses on AMD's architecture, we also provide a comparative evaluation of the features, limitations, and runtime overheads of both Turbo Boost and Turbo CORE technologies. Our results show that we can successful exploit these new hardware facilities to accelerate the execution of key sections of code (critical paths) improving overall performance of some multi-threaded applications. Unlike prior research, we focus on performance instead of power conservation. Our results further can give guidelines for the design of hardware power management facilities and the operating system interfaces to those facilities.

In this thesis a design for a smart high efficiency voltage regulating module capable of supplying the core of modern microprocessors incorporating dynamic voltage and frequency scaling (DVS) capability is accomplished using a RISC based microcontroller to facilitate all the functions required to control, protect, and supply the core with the required variable operating voltage as set by the DVS management system. Normally voltage regulating modules provide maximum power efficiency at designed peak load, and the efficiency falls off as the load moves towards lesser values. A mathematical model has been derived for the main converter and small signal analysis has been performed in order to determine system operation stability and select a control scheme that would improve converter operation response to transients and not requiring intense computational power to realize. A Simulation model was built using Matlab/Simulink and after experimenting with tuned PID controller and fuzzy logic controllers, a simple fuzzy logic control scheme was selected to control the pulse width modulated converter and several methods were devised to reduce the requirements for computational power making the whole system operation realizable using a low power RISC based microcontroller. The same microcontroller provides circuit adaptations operation in addition to providing protection to load in terms of over voltage and over current protection. A novel circuit technique and operation control scheme enables the designed module to selectively change some of the circuit elements in the main pulse width modulated buck converter so as to improve efficiency over a wider range of loads. In case of very light loads as the case when the device goes into standby, sleep or hibernation mode, a secondary converter starts operating and the main converter stops. The secondary converter adapts a different operation scheme using switched capacitor technique which provides high efficiency at low load currents. A fuzzy logic control scheme was chosen for the main converter for its lighter computational power requirement promoting implementation using ultra low power embedded controllers. Passive and active components were carefully selected to augment operational efficiency. These aspects enabled the designed voltage regulating module to operate with efficiency improvement in off peak load region in the range of 3% to 5%. At low loads as the case when the computer system goes to standby or sleep mode, the efficiency improvent is better than 13% which will have noticeable contribution in extending battery run time thus contributing to lowering the carbon footprint of human consumption.

This dissertation proves the feasibility of accurate runtime prediction of processor performance under frequency scaling. The performance predictors developed in this dissertation allow processors capable of dynamic voltage and frequency scaling (DVFS) to improve their performance or energy efficiency by dynamically adapting chip or core voltages and frequencies to workload characteristics. The dissertation considers three processor configurations: the uniprocessor capable of chip-level DVFS, the private cache chip multiprocessor capable of per-core DVFS, and the shared cache chip multiprocessor capable of per-core DVFS. Depending on processor configuration, the presented performance predictors help the processor realize 72–85% of average oracle performance or energy efficiency gains. / text

Les performances des algorithmes d'ordonnancement ont un impact direct sur les performances du système complet. Les algorithmes d'ordonnancement temps réel possèdent des bornes théoriques d'ordonnançabilité optimales mais cette optimalité est souvent atteinte au prix d'un nombre élevé d'événements d'ordonnancement à considérer (préemptions et migrations de tâches) et d'une complexité algorithmique importante. Notre opinion est qu'en exploitant plus efficacement les paramètres des tâches il est possible de rendre ces algorithmes plus efficaces et à coût maitrisé, et ce dans le but d'améliorer la Qualité de Service (QoS) des applications. Nous proposons dans un premier temps des algorithmes d'ordonnancement monoprocesseur qui augmentent la qualité de service d'applications hybrides c'est-à-dire qu'en situation de surcharge, les tâches à contraintes souples ont leur exécution maximisée et les échéances des tâches à contraintes strictes sont garanties. Le coût d'ordonnancement de ces algorithmes est aussi réduit (nombre de préemptions) par une meilleure exploitation des paramètres implicites et explicites des tâches. Cette réduction est bénéfique non seulement pour les performances du système mais elle agit aussi positivement sur la consommation d'énergie. Aussi nous proposons une technique associée à celle de DVFS (dynamic voltage and frequency scaling) afin de minimiser le nombre de changements de points de fonctionnement du fait qu'un changement de fréquence implique un temps d'inactivité du processeur et une consommation d'énergie. Les algorithmes d'ordonnancement multiprocesseur basés sur le modèle d'ordonnancement fluide (notion d'équité) atteignent des bornes d'ordonnançabilité optimales. Cependant cette équité n'est garantie qu'au prix d'hypothèses irréalistes en pratique du fait des nombres très élevés de préemptions et de migrations de tâches qu'ils induisent. Dans cette thèse un algorithme est proposé (ASEDZL) qui n'est pas basé sur le modèle d'ordonnancement fluide. Il permet non seulement de réduire les préemptions et les migrations de tâches mais aussi de relâcher les hypothèses imposées par ce modèle d'ordonnancement. Enfin, nous proposons d'utiliser ASEDZL dans une approche d'ordonnancement hiérarchique ce qui permet d'obtenir de meilleurs résultats que les techniques classiques.

In modern digital audio applications, a continuous audio signal stream is sampled at a fixed sampling rate, which is always greater than twice the highest frequency of the input signal, to prevent aliasing. A more energy efficient approach is to dynamically change the sampling rate based on the input signal. In the dynamic sampling rate technique, fewer samples are processed when there is little frequency content in the samples. The perceived quality of the signal is unchanged in this technique. Processing fewer samples involves less computation work; therefore processor speed and voltage can be reduced. This reduction in processor speed and voltage has been shown to reduce power consumption by up to 40% less than if the audio stream had been run at a fixed sampling rate.

Emergent systems in high-performance computing (HPC) expect maximal efficiency to achieve the goal of power budget under 20-40 megawatts for 1 exaflop set by the Department of Energy. To optimize efficiency, emergent systems provide multiple power-performance control techniques to throttle different system components and scale of concurrency. In this dissertation, we focus on three throttling techniques: CPU dynamic voltage and frequency scaling (DVFS), dynamic memory throttling (DMT), and dynamic concurrency throttling (DCT). We first conduct an empirical analysis of the performance and energy trade-offs of different architectures under the throttling techniques. We show the impact on performance and energy consumption on Intel x86 systems with accelerators of Intel Xeon Phi and a Nvidia general-purpose graphics processing unit (GPGPU). We show the trade-offs and potentials for improving efficiency. Furthermore, we propose a parallel performance model for coordinating DVFS, DMT, and DCT simultaneously. We present a multivariate linear regression-based approach to approximate the impact of DVFS, DMT, and DCT on performance for performance prediction. Validation using 19 HPC applications/kernels on two architectures (i.e., Intel x86 and IBM BG/Q) shows up to 7% and 17% prediction error correspondingly. Thereafter, we develop the metrics for capturing the performance impact of DVFS, DMT, and DCT. We apply the artificial neural network model to approximate the nonlinear effects on performance impact and present a runtime control strategy accordingly for power capping. Our validation using 37 HPC applications/kernels shows up to a 20% performance improvement under a given power budget compared with the Intel RAPL-based method. / Ph. D. / System efficiency on high-performance computing (HPC) systems is the key to achieving the goal of power budget for exascale supercomputers. Techniques for adjusting the performance of different system components can help accomplish this goal by dynamically controlling system performance according to application behaviors. In this dissertation, we focus on three techniques: adjusting CPU performance, memory performance, and the number of threads for running parallel applications. First, we profile the performance and energy consumption of different HPC applications on both Intel systems with accelerators and IBM BG/Q systems. We explore the trade-offs of performance and energy under these techniques and provide optimization insights. Furthermore, we propose a parallel performance model that can accurately capture the impact of these techniques on performance in terms of job completion time. We present an approximation approach for performance prediction. The approximation has up to 7% and 17% prediction error on Intel x86 and IBM BG/Q systems respectively under 19 HPC applications. Thereafter, we apply the performance model in a runtime system design for improving performance under a given power budget. Our runtime strategy achieves up to 20% performance improvement to the baseline method.

Current Intel processors implement a variety of power saving features like frequency scaling and idle states. These mechanisms limit the power draw and thereby decrease the thermal dissipation of the processors. However, they also have an impact on the achievable performance. The various mechanisms significantly differ regarding the amount of power savings, the latency of mode changes, and the associated overhead. In this paper, we describe and closely examine the so-called software controlled clock modulation mechanism for different processor generations. We present results that imply that the available documentation is not always correct and describe when this feature can be used to improve energy efficiency. We additionally compare it against the more popular feature of dynamic voltage and frequency scaling and develop a model to decide which feature should be used to optimize inter-process synchronizations on Intel Haswell-EP processors.

Current Intel processors implement a variety of power saving features like frequency scaling and idle states. These mechanisms limit the power draw and thereby decrease the thermal dissipation of the processors. However, they also have an impact on the achievable performance. The various mechanisms significantly differ regarding the amount of power savings, the latency of mode changes, and the associated overhead. In this paper, we describe and closely examine the so-called software controlled clock modulation mechanism for different processor generations. We present results that imply that the available documentation is not always correct and describe when this feature can be used to improve energy efficiency. We additionally compare it against the more popular feature of dynamic voltage and frequency scaling and develop a model to decide which feature should be used to optimize inter-process synchronizations on Intel Haswell-EP processors.

Several modern multi-core architectures support the dynamic control of the CPU's clock rate, allowing processor cores to temporarily operate at speeds exceeding the operational base frequency. Conversely, cores can operate at a lower speed or be disabled altogether to save power. Such facilities are notably provided by Intel's Turbo Boost and AMD's Turbo CORE technologies. Frequency control is typically driven by the operating system which requests changes to the performance state of the processor based on the current load of the system. In this paper, we investigate the use of dynamic frequency scaling from user space to speed up multi-threaded applications that must occasionally execute time-critical tasks or to solve problems that have heterogeneous computing requirements. We propose a general-purpose library that allows selective control of the frequency of the cores - subject to the limitations of the target architecture. We analyze the performance trade-offs and illustrate its benefits using several benchmarks and real-world workloads when temporarily boosting selected cores executing time-critical operations. While our study primarily focuses on AMD's architecture, we also provide a comparative evaluation of the features, limitations, and runtime overheads of both Turbo Boost and Turbo CORE technologies. Our results show that we can successful exploit these new hardware facilities to accelerate the execution of key sections of code (critical paths) improving overall performance of some multi-threaded applications. Unlike prior research, we focus on performance instead of power conservation. Our results further can give guidelines for the design of hardware power management facilities and the operating system interfaces to those facilities.