Dynamic Voltage And Frequency Scaling Research Articles

A holistic energy-efficient approach for a processor-memory system

Component overclocking is an effective approach to speed up the components of a system to realize a higher program performance; it includes processor overclocking or memory overclocking. However, overclocking will unavoidably result in increase in power consumption. Our goal is to optimally improve the performance of scientific computing applications without increasing the total power consumption for a processor-memory system. We built a processor-memory energy efficiency model for multicore-based systems, which coordinates the performance and power of processor and memory. Our model exploits performance boost opportunities for a processor-memory system by adopting processor overclocking, processor Dynamic Voltage and Frequency Scaling (DVFS), memory active ratio adjustment, and memory overclocking, according to different scientific applications. This model also provides a total power control method by considering the same four factors mentioned above. We propose a processor and memory Coordination-based holistic Energy-Efficient (CEE) algorithm, which achieves performance improvement without increasing the total power consumption. The experimental results show that an average of 9.3% performance improvement was obtained for all 14 benchmarks. Meanwhile the total power consumption does not increase. The maximal performance improvement was up to 13.1% from dedup benchmark. Our experiments validate the effectiveness of our holistic energy-efficient model and technology.

Adaptive energy-aware scheduling method in a meteorological cloud

This paper focuses on the energy-aware scheduling problem of moldable non-linear parallel tasks in a meteorological cloud. Such a meteorological Cloud mainly provides computing resources for the execution of meteorological models, such as Weather Research and Forecasting model (WRF). In a meteorological Cloud, the parallelism of tasks (i.e., meteorological models) can only be configured in the beginning, and the assigned resources retained exclusively until all sub-tasks have been finished. For the scheduling of those tasks, one key challenge is how to reduce the average energy consumption while guaranteeing others requirements of such tasks. We address this challenge by considering simultaneously the deadlines of tasks, the energy consumption, the system load, and the non-linear speedup of parallel tasks when we make the scheduling decision. Specifically, we propose an adaptive energy-aware scheduling method called ASSD, that is based on the Dynamic Voltage and Frequency Scaling (DVFS) model of computing resources and the speedup of tasks under different parallelisms. We evaluate our method via simulations on a meteorological cloud. Our results show that the proposed method not only increases the number of completed tasks but also significantly reduces the average energy consumption.

Energy-aware scheduling using slack reclamation for cluster systems

Reducing energy consumption in processors has become a critical issue in cluster systems. The dynamic voltage and frequency scaling (DVFS) is one of the effective techniques to minimize energy consumption in processors when executing parallel applications. These applications may consist of many interrelated tasks that may be computed when their precedence constraints are satisfied. The scheduling of these precedence-constrained tasks on to processors, to optimize makespan or energy consumption or both, has been studied as an NP-complete problem. Thus, many algorithms have been given by the researchers and newer algorithms keep coming in. In the parallel applications, there exist some tasks which may have slacks due to the dependencies between tasks. The EASLA (energy aware service level agreement) is one of the well-known energy-aware task scheduling algorithms which utilize the underused slack of tasks. In this paper, we present an improved version of its sub-algorithm, NCM (Not Changing Makespan), for heterogeneous cluster systems that is used to downscale frequencies when schedule length does not change. The presented algorithm also makes use of a fast and low complexity algorithm called PEFT (Predict Earliest Finish Time) to compute the schedule length of the application. We performed experiments for randomly generated graphs, and the results illustrate that the improved version of the NCM algorithm achieves better energy saving compared to the NCM algorithm.

HEALERS: a heterogeneous energy‐aware low‐overhead real‐time scheduler

Devising energy-efficient scheduling strategies for real-time periodic tasks on heterogeneous platforms is a challenging as well as a computationally demanding problem. This study proposes a low-overhead heuristic strategy called, HEALERS, for dynamic voltage and frequency scaling (DVFS)-cum-dynamic power management (DPM) enabled energy-aware scheduling of a set of periodic tasks executing on a heterogeneous multi-core system. The presented strategy first applies deadline-partitioning to acquire a set of distinct time-slices. At any time-slice boundary, the following three-phase operations are applied to obtain a schedule for the next time-slice: first, it computes the fragments of the execution demands of all tasks onto each of the different processing cores in the platform. Next, it generates a schedule for each task on one or more processing cores such that the total execution demand of all tasks is satisfied. Finally, HEALERS applies DVFS and DPM on all processing cores so that energy consumption within the time-slice may be minimized while not jeopardising execution requirements of the scheduled tasks. Experimental results show that the proposed scheme is not only able to achieve appreciable energy savings with respect to state-of-the-art (5–42% on average) but also enables a significant improvement in resource utilisation (as high as 58%).

Dynamic Power Management for Neuromorphic Many-Core Systems

This paper presents a dynamic power management architecture for neuromorphic many-core systems, such as SpiNNaker. A fast dynamic voltage and frequency scaling (DVFS) technique is presented which allows the processing elements (PEs) to change their supply voltage and clock frequency individually and autonomously within less than 100 ns. This is employed by the neuromorphic simulation software flow, which defines the performance level (PL) of the PE based on the actual workload within each simulation cycle. A test chip in 28-nm SLP CMOS technology has been implemented. It includes four PEs which can be scaled from 0.7 to 1.0 V with frequencies from 125 to 500 MHz at three distinct PLs. By measurement of three neuromorphic benchmarks, it is shown that the total PE power consumption can be reduced by 75%, with 80% baseline power reduction and a 50% reduction of energy per neuron and synapse computation, all while maintaining temporary peak system performance to achieve biological real-time operation of the system. A numerical model of this power management model is derived which allows DVFS architecture exploration for neuromorphics. The proposed technique is to be used for the second-generation SpiNNaker neuromorphic many-core system.

Lifetime-aware real-time task scheduling on fault-tolerant mixed-criticality embedded systems

Abstract In recent years, the design of mixed-criticality embedded systems suffering from transient faults has attracted much attention. From the perspective of system users, it is desirable to optimize system lifetime while meeting all design constraints. Existing task scheduling algorithms cannot be utilized to maximize the lifetime of mixed-criticality embedded systems since they do not take into account the impact of providing transient fault tolerance on system lifetime. This paper investigates the problem of prolonging the lifetime of mixed-criticality embedded systems on a uniprocessor equipped with dynamic voltage and frequency scaling (DVFS) technique. The transient faults and thermal cycling incurred permanent faults are simultaneously considered in the system lifetime optimization under the constraints of safety requirements and schedule timeliness. A mixed-integer linear programming (MILP) formulation is first presented to deal with the task scheduling problem. Since the MILP method is a time-consuming solution for large-scale systems, a cross-entropy method based heuristic is then proposed to achieve a better tradeoff between the system lifetime achieved by the derived task schedule and the runtime consumed to generate the task schedule. Experiments based on synthetic and real-world benchmarks are conducted, and simulation results demonstrate that the proposed heuristic improves system lifetime by up to 32.73% with acceptable runtime as compared to benchmarking methods.

GDP: A Greedy Based Dynamic Power Budgeting Method for Multi/Many-Core Systems in Dark Silicon

Dark silicon phenomenon is significant in today's multi/many-core systems manufactured using new generation technology. In order to enhance performance of dark silicon systems, power budget constrained dynamic optimizations are performed in various ways including dynamic voltage and frequency scaling (DVFS) and task scheduling. However, power budgets given by existing methods are generally over pessimistic, which greatly limit the capability of dynamic performance optimization methods. In order to resolve this problem, we propose a dynamic power budgeting method, called Greedy based Dynamic Power (GDP). Different from existing methods, which are steady state based and ignore active core distributions, GDP formulates the power budgeting problem as a thermal-constrained combinational power optimization problem. To efficiently solve this problem, we propose two new ideas: first, we transform the original power-optimization problem to an easier solving temperature-optimization problem; second, we employ a more efficient greedy based algorithm that finds a sub-optimal active core distribution which maximizes power budget. The new method can consider current temperature states and transient thermal effects, which were ignored by existing methods. Both theoretical studies and experimental results show that GDP outperforms existing methods by providing a higher and less pessimistic power budget with low computing cost and guaranteed thermal safety.

Improving the energy efficiency of relational and NoSQL databases via query optimizations

As big data becomes the norm of various industrial applications, the complexity of database workloads and database system design has increased significantly. To address these challenges, conventional relational databases have been constantly improved and NoSQL databases such as MongoDB and Cassandra have been proposed and implemented to compete with SQL databases. In addition to traditional metrics such as response time, throughput, and capacity, modern database systems are posing higher requirements on energy efficiency due to the large volume of data that need to be stored, queried, updated, and analyzed. While decades of research in the database and data processing communities has produced a wealth of literature that optimize for performance, research on optimizations for energy efficiency has been historically overlooked and only a few studies have investigated the energy efficiency of database systems. To the best our knowledge, there are currently no comprehensive studies that analyze the impact of query optimizations on performance and energy efficiency across both relational and NoSQL databases. In fact, the energy behavior of many basic database operations (e.g. insertion, deletion, searching, update, indexing, etc) remains largely unknown due to the lack of accurate power measurement methodologies for various databases and queries. In this paper, we investigate a series of query optimization techniques for improving the energy-efficiency of relational databases and NoSQL databases. We use both widely acceptable benchmarks (e.g. Yahoo! Cloud Server Benchmark) and customized datasets (converted from ˜100GB of Twitter data) in our experiments to evaluate the effectiveness of various optimization techniques. We conduct cross database analysis on relational database (MySQL) and NoSQL based databases (MongoDB and Cassandra) to compare their performance and energy efficiency. Additionally, we study a variety of optimization techniques that can improve energy efficiency without compromising performance on the databases derived from the Twitter data. Using these techniques, we are able to achieve significant energy savings without performance degradation. Moreover, we investigate the impact of Dynamic Voltage and Frequency Scaling (DVFS) on the performance and energy efficiency of MySQL, MongoDB and Cassandra.

Dynamic Voltage and Frequency Scaling in NoCs with Supervised and Reinforcement Learning Techniques

Network-on-Chips (NoCs) are the de facto choice for designing the interconnect fabric in multicore chips due to their regularity, efficiency, simplicity, and scalability. However, NoC suffers from excessive static power and dynamic energy due to transistor leakage current and data movement between the cores and caches. Power consumption issues are only exacerbated by ever decreasing technology sizes. Dynamic Voltage and Frequency Scaling (DVFS) is one technique that seeks to reduce dynamic energy; however this often occurs at the expense of performance. In this paper, we propose LEAD Learning-enabled Energy-Aware Dynamic voltage/frequency scaling for multicore architectures using both supervised learning and reinforcement learning approaches. LEAD groups the router and its outgoing links into the same V/F domain and implements proactive DVFS mode management strategies that rely on offline trained machine learning models in order to provide optimal V/F mode selection between different voltage/frequency pairs. We present three supervised learning versions of LEAD that are based on buffer utilization, change in buffer utilization and change in energy/throughput, which allow proactive mode selection based on accurate prediction of future network parameters. We then describe a reinforcement learning approach to LEAD that optimizes the DVFS mode selection directly, obviating the need for label and threshold engineering. Simulation results using PARSEC and Splash-2 benchmarks on a 4 × 4 concentrated mesh architecture show that by using supervised learning LEAD can achieve an average dynamic energy savings of 15.4 percent for a loss in throughput of 0.8 percent with no significant impact on latency. When reinforcement learning is used, LEAD increases average dynamic energy savings to 20.3 percent at the cost of a 1.5 percent decrease in throughput and a 1.7 percent increase in latency. Overall, the more flexible reinforcement learning approach enables learning an optimal behavior for a wider range of load environments under any desired energy versus throughput tradeoff.

Model-based optimization of the energy efficiency of multi-threaded applications

Energy efficiency is considered to be a critical concern for modern hardware and a variety of hardware features have been developed to improve the energy balance for executing applications. This article focuses on the dynamic voltage and frequency scaling (DVFS) technique, which is available for many platforms. Analytical models for capturing the energy efficiency are considered and it is investigated whether such an analytical model is able to support an a priori selection of the operational frequency that leads to a near optimal energy consumption for the application code to be executed. Furthermore, the possible influence of the number of threads executing a multi-threaded application is taken into account. Models for the energy-delay product (EDP) weighting the power against the square of execution time are also covered. The experimental evaluation is performed on the basis of the multi-threaded PARSEC and SPLASH-2 benchmarks executed on four different recent multi-cores. The investigations show that the operational frequency selected according to the analytical models leads to an energy consumption that is near the minimum energy consumption over all frequencies available.

An energy-efficient, QoS-aware and cost-effective scheduling approach for real-time workflow applications in cloud computing systems utilizing DVFS and approximate computations

Green cloud computing attracts significant attention from both academia and industry. One of the major challenges involved, is to provide a high level of Quality of Service (QoS) in a cost-effective way for the end users and in an energy-efficient manner for the cloud providers. Towards this direction, this paper presents an energy-efficient, QoS-aware and cost-effective scheduling strategy for real-time workflow applications in cloud computing systems. The proposed approach utilizes per-core Dynamic Voltage and Frequency Scaling (DVFS) on the underlying heterogeneous multi-core processors, as well as approximate computations, in order to fill in schedule gaps. At the same time, it takes into account the effects of input error on the processing time of the component tasks. Our goal is to provide timeliness and energy efficiency by trading off result precision, while keeping the result quality of the completed jobs at an acceptable standard and the monetary cost required for the execution of the jobs at a reasonable level. The proposed scheduling heuristic is compared to two other baseline policies, under the impact of various QoS requirements. The simulation experiments reveal that our approach outperforms the other examined policies, providing promising results.

Reliability and energy efficient workflow scheduling in cloud environment

Cloud data centers consume huge amounts of electrical energy which results in an increased operational cost, decreased system reliability and carbon dioxide footprints. Thus, it is highly important to develop scheduling strategy to reduce energy consumption. Dynamic voltage and frequency scaling (DVFS) has been recognized as an efficient technique for reducing energy consumption. However, there is negative impact of DVFS on the reliability of system as it increases the transient faults during the application execution. Hence, it is essential to address the issue of reliability for mission critical applications. Recent studies on workflow scheduling in distributed environment have not considered reliability while minimizing the energy consumption. In this paper, we propose a new scheduling algorithm called the reliability and energy efficient workflow scheduling algorithm which jointly optimizes lifetime reliability of application and energy consumption and guarantees the user specified QoS constraint. The proposed algorithm works in four phases: priority calculation, clustering of tasks, distribution of target time and assigning the cluster to processing element with appropriate voltage/frequency levels. The simulation results obtained by using randomly generated task graphs and Gaussian Elimination task graphs shows that the proposed approach is effective in joint optimization of lifetime reliability of system and energy consumption compared to existing algorithms.

GPU Energy Consumption Optimization With a Global-Based Neural Network Method

With the widespread use of smart technologies, graphics processing unit (GPU) power-optimization issues are becoming increasingly important. Many researchers have tried to use dynamic voltage and frequency scaling (DVFS) technology to optimize a GPU’s internal energy consumption. However, DVFS energy management often has difficulty balancing GPU performance and energy efficiency. This paper aims to implement a DVFS energy management strategy. We constructed a new type of neural network to a GPU-based energy management scheme, implemented the global-based DVFS model, and explored its implementation details. Using a master-slave model, we built a global energy control solution strategy. This strategy performs global collaborative DVFS adjustments on the GPU’s energy consumption module based on task characteristics. Through the software construction and implementation of the global-based DVFS model, we proved that the strategy improves the GPU performance while improving the GPU’s energy efficiency. We conducted performance and energy tests on three GPUs on the Tesla, Fermi, and Kepler platforms. The experiments showed that this strategy improved the performance and power consumption of GPUs based on each of the platforms.

DPVFS: a dynamic procrastination cum DVFS scheduler for multi-core hard real-time systems

Optimising energy consumption has become the primary focus of research in recent years. Static and dynamic energy optimisation during task scheduling is one of the most prominent measures available. This is achieved mainly by shutdown and slowdown techniques. In uni-processor real-time systems, the most widely used shutdown and slowdown techniques are dynamic procrastination (DP) and dynamic voltage and frequency scaling (DVFS). This paper proposes DPVFS a hard real-time task scheduler for multi-core (MC) system to optimise overall energy consumption without deadline misses. DPVFS combines DP and DVFS for MC systems to save overall energy consumption. DPVFS shuts the processor down whenever possible with the help of procrastination. If shutdown is not possible, it adjusts the voltage and frequency to reduce dynamic energy consumption. The experimental evaluation of DPVFS with synthetically generated benchmark program suites shows savings of 18.8% and 33.2% of overall energy over DP based schedulers and DVFS based schedulers respectively.

Energy oriented EDF for real-time systems

Energy is a major concern when designing real-time systems. A common method for saving energy while still guaranteeing the real-time constraints is to embed dynamic voltage and frequency scaling (DVFS) mechanisms and dynamic power management (DPM) mechanisms within a real-time scheduling algorithm such as EDF. This paper proposes a new extension to the EDF scheduler, termed energy oriented EDF (EO-EDF). The new scheduler makes it possible to change the original EDF task execution order to better utilise the slack time and thus decrease the energy consumption, while still meeting the task deadlines. The new task order is defined according to a novel criterion we invented, termed task prediction order (TPO). The paper introduces two new versions of the EO-EDF scheduler, termed TPO-EDF and STPO-EDF. While STPO-EDF applies the TPO criterion in a static manner, TPO-EDF allows it to be used dynamically. We simulate the new proposed algorithms using both synthetic workloads and real-time benchmarks. The evaluations show that integrating both the TPO-EDF and STPO-EDF scheduling algorithms with DVFS and DPM mechanisms achieves an energy savings of 30% on average, in comparison with current known EDF based scheduling utilising DVFS and DPM mechanisms.

PROGRES: A Programmable Green Router With Controlled Service Rate

Routers are the most common transmission devices over the Internet. Architecture of the routers can be improved to fully utilize the chances to save energy related to the CMOS based transmission equipment. One prime approach for controlling and managing the power dissipation of the CMOS devices is Dynamic Voltage and Frequency Scaling (DVFS). This paper implements the DVFS in a CMOS device programmed as a router for studying multiple factors impeding energy saving by using Dynamic Clock Generation (DCG) and Dynamic Clock Selection (DCS). The authors emphasize on the performance evaluation of such frequency scaled CMOS devices, in which the service rate vary according to the state of an underlying finite-state, continuous-time Markov model. Furthermore, this paper establishes the relationships of critical factors (in/)directly effecting power consumption of these devices. Aforementioned factors and their dependencies such as the arrival and service rates, queuing lengths and thresholds in such systems facilitate the changing dynamics of functions and services for appropriate clock state of the device. Thus, the resulting model is quite complex to be proved analytically or numerically. We investigate and implement PROGRES to affirm the numerical models' accuracy for capturing the service rate transitions of the proposed control policies with detailed performance analysis and discussion. Furthermore, PROGRES also scales the arrival rates and queue lengths, while ensuring the controlled transition rates of the modulating Markov process in the programmable router. Two control models against single service rate for benchmarking. These techniques and numerical experiments allow us to identify and evaluate the trade-offs between energy consumption and system performance in the futuristic programmable networking devices with resulting approximations are promising at network, node and functional levels.

Energy-Efficient Scheduling for Real-Time Systems Based on Deep Q-Learning Model

Energy saving is a critical and challenging issue for real-time systems in embedded devices because of their limited energy supply. To reduce the energy consumption, a hybrid dynamic voltage and frequency scaling (DVFS) scheduling based on Q-learning (QL-HDS) was proposed by combining energy-efficient DVFS techniques. However, QL-HDS discretizes the system state parameters with a certain step size, resulting in a poor distinction of the system states. More importantly, it is difficult for QL-HDS to learn a system for various task sets with a Q-table and limited training sets. In this paper, an energy-efficient scheduling scheme based on deep Q-learning model is proposed for periodic tasks in real-time systems (DQL-EES). Specially, a deep Q-learning model is designed by combining a stacked auto-encoder and a Q-learning model. In the deep Q-learning model, the stacked auto-encoder is used to replace the Q-function for learning the Q-value of each DVFS technology for any system state. Furthermore, a training strategy is devised to learn the parameters of the deep Q-learning model based on the experience replay scheme. Finally, the performance of the proposed scheme is evaluated by comparison with QL-HDS on different simulation task sets. Results demonstrated that the proposed algorithm can save average <inline-formula><tex-math notation="LaTeX">$4.2\%$</tex-math></inline-formula> energy than QL-HDS.

GPU Static Modeling Using PTX and Deep Structured Learning

In the quest for exascale computing, energy-efficiency is a fundamental goal in high-performance computing systems, typically achieved via dynamic voltage and frequency scaling (DVFS). However, this type of mechanism relies on having accurate methods of predicting the performance and power/energy consumption of such systems. Unlike previous works in the literature, this research focuses on creating novel GPU predictive models that do not require run-time information from the applications. The proposed models, implemented using recurrent neural networks, take into account the sequence of GPU assembly instructions (PTX) and can accurately predict changes in the execution time, power and energy consumption of applications when the frequencies of different GPU domains (core and memory) are scaled. Validated with 24 applications on GPUs from different NVIDIA microarchitectures (Turing, Volta, Pascal and Maxwell), the proposed models attain a significant accuracy. Particularly, the obtained power consumption scaling model provides an average error rate of 7.9% (Tesla T4), 6.7% (Titan V), 5.9% (Titan Xp) and 5.4% (GTX Titan X), which is comparable to state-of-the-art run-time counter-based models. When using the models to select the minimum-energy frequency configuration, significant energy savings can be attained: 8.0% (Tesla T4), 6.0% (Titan V), 29.0% (Titan Xp) and 11.5% (GTX Titan X).

Frequency Selection Approach for Energy Aware Cloud Database

A lot of cloud systems are adopted in industry and academia to face the explosion of the data volume and the arrival of the big data era. Meanwhile, energy efficiency and energy saving become major concerns for data centers where massive cloud systems are deployed. However, energy waste is quite common due to resource provisioning. In this paper, using dynamic voltage and frequency scaling (DVFS), a frequency selection approach is introduced to improve the energy efficiency of the cloud systems in terms of resource provisioning. In the approach, two algorithms, genetic algorithm (GA) and Monte Carlo tree search algorithm (MCTS), are proposed. A cloud database system is taken as an example to evaluate the approach. The results of the experiments show that both algorithms have its advantages. The algorithms have great scalability, in which the GA can be applied to thousands of nodes and the MCTS can be applied to hundreds of nodes. Both algorithms have high accuracy compared with optimal solutions (up to 99.9% and 99.6% for GA and MCTS, respectively). According to an optimality bound analysis, 26% of energy can be saved at most using our frequency selection approach.

Energy-Efficient Intra-task DVFS Scheduling Using Linear Programming Formulation

In real-time embedded systems, minimizing energy consumption is one of the most important tasks. Intra-task dynamic voltage and frequency scaling (DVFS) has been the subject of much research in the task boundary of time-constrained applications for energy reduction. The problem of optimizing energy consumption with respect to intra-task DVFS scheduling can be addressed by assigning proper operational frequencies to individual basic blocks in a program while guaranteeing the deadline. Based on the profile information of a task, we first formulate the problem in terms of integer linear programming (ILP) regarding different assumptions of transition overhead. To verify the effectiveness of ILP formulations, the most representative intra-task DVFS techniques are taken for comparisons. The results of the experiments demonstrate that the proposed ILP method achieves greater energy savings than the existing approaches. Moreover, it determines the optimal scheduling strategy in reasonable execution time for applications with a limited number of blocks.