Transition Overhead Research Articles

Energy-aware fixed priority scheduling with shared resources in standby-sparing systems

The problem of joint energy and reliability management for fixed priority periodic real-time tasks set with shared resources on a standby-sparing system is studied in this paper. A novel algorithm called FPMPSA is proposed. It considers frequency transition and processor state transition overheads imposed by dynamic voltage frequency scaling (DVFS) and dynamic power management (DPM). In addition, it applies the rate monotonic with dual priority scheduling policy (RM/DPP) to ensure that shared resources can be accessed in a mutual exclusive manner. Moreover, the mixed mapping method is applied to map the main task and backup task. Furthermore, the energy efficient speed of the main task SM is determined and the backup task executes with the maximum processor speed. For energy efficiency, the algorithm called DFPMPSA which is an extension of FPMPSA is proposed. It can use the dynamic slack time to reduce energy consumption. The experimental results show that DFPMPSA algorithm consumes average 8.15% less energy than that of FPMPSA.

COUNTDOWN: A Run-Time Library for Performance-Neutral Energy Saving in MPI Applications

Power and energy consumption are becoming key challenges for the supercomputers’ exascale race. HPC systems’ processors waist active power during communication and synchronization among the MPI processes in large-scale HPC applications. However, due to the time scale at which communication happens, transitioning into low-power states while waiting for the completion of each communication may introduce unacceptable overhead. In this article, we present COUNTDOWN, a run-time library for identifying and automatically reducing the power consumption of the CPUs during communication and synchronization. COUNTDOWN saves energy without penalizing the time-to-completion by lowering CPUs power consumption only during idle times for which power state transition overhead is negligible. This is done transparently to the user, without requiring labor-intensive and error-prone application code modifications, nor requiring recompilation of the application. We test our methodology on a production Tier-1 system. For the NAS benchmarks, COUNTDOWN saves between 6 and 50 percent energy, with a time-to-solution penalty lower than 5 percent. In a complete production—Quantum ESPRESSO—for a 3.5K cores run, COUNTDOWN saves 22.36 percent energy, with a performance penalty below 3 percent. Energy saving increases to 37 percent with a performance penalty of 6.38 percent, if the application is executed without communication tuning.

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.

Deep Learning-Based Sustainable Data Center Energy Cost Minimization With Temporal MACRO/MICRO Scale Management

Recently, distributed sustainable data centers based on renewable power generators have been deployed in order to efficiently reduce both the energy cost and carbon emission. Dynamic right-sizing (DRS) and frequency scaling (FS) have been considered as promising solutions to tune the computing capacity corresponding to the dynamic renewable power capacity. However, in existing works, the inaccurate power prediction and the uncoordinated decision making of DRS/FS still lead to low service quality and high energy cost. In this paper, we propose a novel joint optimization method for energy efficient distributed sustainable data centers. The proposed method adopts long short-term memory approach to improve the prediction accuracy of renewable power capacity for a long period, and unsupervised deep learning (DL) solver to resolve the coordinated DRS/FS optimization. Furthermore, we present the MACRO/MICRO (MAMI) time scale-based data center management technique to achieve both high energy efficiency and low wake-up transition overhead of DRS. To evaluate the proposed DL-based MAMI optimizer, we use the real trace data of renewable power capacity from the U.S. Measurement and Instrumentation Data Center. The experimental results demonstrate that our method reduces the energy cost by 25% compared with conventional metaheuristics while guaranteeing the service response delay requirement.

Race to idle or not: balancing the memory sleep time with DVS for energy minimization

Reducing energy consumption is a critical problem in most of the computing systems today. In recent years, dynamic voltage scaling (DVS) has been often applied in the multi-core processor systems. The leakage power of the main memory shared by the multiple DVS cores is becoming a larger problem with technology scaling. This paper focuses on minimizing the system-wide energy consumption by applying DVS on each core and turning the memory to sleep when all the cores have common idle time. This work presents systematic analysis for the target problem based on different system models and task models. For tasks with common release time, optimal schemes are presented for the systems both with and without considering the static power of the cores. For the general task model, a heuristic online algorithm is proposed. Furthermore, the scheme is extended to handle the problem when the transition overhead between the active and sleep modes is not negligible. The experimental results show that the heuristic algorithm can reduce the energy consumption of the overall system by 8.73% in average (up to 28.44%) compared to a state-of-the-art multi-core DVS scheduling scheme.

Energy‐Aware Scheduling with Uncertain Execution Time for Real‐Time Systems

Energy-saving is extraordinary important for real-time systems. Dynamic voltage scaling (DVS) is an important technique to reduce the energy consumption of processors that support voltage scaling. It has been exploited extensively in task scheduling. However, many approaches take simple treatments and some of them even neglect the large voltage transition overheads. Although some strategies consider the penalty, frequency switching produces extra large time overhead and deadline misses occur frequently. In our paper, we propose an energy efficient soft real-time dynamic program scheme by using quantitative switching overhead and communication penalty for multitask scheduling with uncertain execution time. The experiments show that our approaches significantly outperform existing solutions both on simple-core and multi-core systems in terms of energy-saving.

Maximizing power saving with state transition overhead for multiple mobile subscriber stations in WiMAX

In the IEEE 802.16e/m standard, three power saving classes (PSCs) are defined to save the energy of a mobile sub-scriber station (MSS). However, how to set the parameters of PSCs to maximize the power saving and guarantee the quality of service is not specified in the standard. Thus, many algorithms were proposed to set the PSCs in IEEE 802.16 networks. However, most of the proposed algorithms consider only the power saving for a single MSS. In the algorithms designed for multiple MSSs, the sleep state, which is set for activation of state transition overhead power, is not considered. The PSC setting for real-time connections in multiple MSSs with consideration of the state transition overhead is studied. The problem is non-deterministic polynomial time hard (NP-hard), and a suboptimal algorithm for the problem is proposed. Simulation results demonstrate that the energy saving of the proposed algorithm is higher than that of state-of-the-art algorithms and approaches the optimum limit.

Low power fixed priority scheduling sporadic task with shared resources in hard real time systems

Dynamic voltage scaling (DVS) and dynamic power management (DPM) are two effective techniques in a real time system. In this paper, we address the problem of the canonical sporadic task scheduling based on a fixed-priority scheduling scheme and take a generalized power model into account. The sporadic tasks share a set serially reusable, single-unit resources. First, we present a rate monotonic with dual priority scheduling policy, called RM/DPP, to solve the sporadic tasks shared resources scheduling problem and discuss the feasibility of the RM/DPP algorithm. Second, a static fixed-priority sporadic tasks scheduling algorithm with shared resources, called SFPSASR, has been put forward, which considers the off-chip workload and assumes that each task executes with its worst case execution time. Third, for energy efficiency, a dynamic fixed-priority sporadic tasks scheduling algorithm with shared resources, called DFPSASR, has been put forward, which considers the speed transition overhead and combines the DVS and the DPM technology. The experimental results show that the proposed SFPSASR algorithm can reduce the energy consumption by 42.14%∼51.73% over the RM/DPP algorithm and the DFPSASR algorithm can reduce the energy consumption by 79.37%∼82.94% over the SFPSASR algorithm.

인터럽트 병합 최적화를 통한 네트워크 장치 에너지 절감 방법 연구

The concept of energy-efficient networking has begun to spread in the past few years, gaining increasing popularity. A common opinion among networking researchers is that the sole introduction of low consumption silicon technologies may not be enough to effectively curb energy requirements. Thus, for disruptively boosting the network energy efficiency, these hardware enhancements must be integrated with ad-hoc mechanisms that explicitly manage energy saving, by exploiting network-specific features. The IEEE 802.3az Energy Efficient Ethernet (EEE) standard is one of such efforts. EEE introduces a low power mode for the most common Ethernet physical layer standards and is expected to provide large energy savings. However, it has been shown that EEE may not achieve good energy efficiency because mode transition overheads can be significant, leading to almost full energy consumption even at low utilization levels. Coalescing techniques such as packet coalescing and interrupt coalescing were proposed to improve energy efficiency of EEE, but their implementations typically adopt a simple policy that employs a few fixed values for coalescing parameters, thus it is difficult to achieve optimal energy efficiency. The paper proposes adaptive interrupt coalescing (AIC) that adopts an optimal policy that could not only improve energy efficiency but support performance. AIC has been implemented at the sender side with the Intel 82579 network interface card (NIC) and e1000e Linux device driver. The experiments were performed at 100 M bps transfer rate and show that energy efficiency of AIC is improved in most cases despite performance consideration and in the best case can be improved up to 37% compared to that of conventional interrupt coalescing techniques.

Transition-aware task scheduling and configuration selection in reconfigurable embedded systems

This paper presents a novel graph representation that captures the transition overhead due to runtime configuration of underlying hardware in reconfigurable embedded systems resulting from various configuration schemes such as FPGA-like reconfiguration or dynamic voltage/frequency scaling (DVFS). We propose an intuitive heuristic to solve combined configuration selection and task scheduling problem on this graph. In addition, when applied to DVFS, our algorithm provides simultaneous task ordering and configuration selection of the system which outperforms the state-of-the-art DVFS methods applied after task ordering.

Energy and transition-aware runtime task scheduling for multicore processors

Many embedded or portable devices have large demands on running real-time applications. The designers start to adopt the multicore processors in these devices. The multi-core processors, however, cause much higher power consumption than ever before. To resolve this problem, many researchers have focused their studies on designing the energy-aware task scheduling algorithms for multicore processors. Conventional scheduling algorithms assumed that each core can operate under different voltage levels. However, they have not considered the effects of voltage transition overheads, which may defeat the benefit of task scheduling. In this paper, we aim to resolve this scheduling problem with voltage transition overhead consideration. We formalize this problem by an integer linear programming model and propose a heuristic algorithm for a runtime environment. The experimental results show that the proposed online heuristic algorithm can obtain the comparable results with the optimal scheduling derived by the offline integer linear programming approach.

Accurate Modeling of the Delay and Energy Overhead of Dynamic Voltage and Frequency Scaling in Modern Microprocessors

Dynamic voltage and frequency scaling (DVFS) has been studied for well over a decade. Nevertheless, existing DVFS transition overhead models suffer from significant inaccuracies; for example, by incorrectly accounting for the effect of DC–DC converters, frequency synthesizers, voltage, and frequency change policies on energy losses incurred during mode transitions. Incorrect and/or inaccurate DVFS transition overhead models prevent one from determining the precise break-even time and thus forfeit some of the energy saving that is ideally achievable. This paper introduces accurate DVFS transition overhead models for both energy consumption and delay. In particular, we redefine the DVFS transition overhead including the underclocking-related losses in a DVFS-enabled microprocessor, additional inductor IR losses, and power losses due to discontinuous-mode DC–DC conversion. We report the transition overheads for a desktop, a mobile and a low-power representative processor. We also present DVFS transition overhead macromodel for use by high-level DVFS schedulers.

Discrete feedback-based dynamic voltage scaling for safety critical real-time systems

Recently, the tradeoff between low energy consumption and high fault-tolerance has attracted a lot of attention as a key issue in the design of real-time systems. Dynamic Voltage Scaling (DVS) is commonly employed as one of the most effective low energy techniques for real-time systems. It has been observed that the use of feedback-based methods can improve the effectiveness of DVS-enabled systems. In this paper, we have investigated reducing the energy consumption of fault-tolerant hard real-time systems using the feedback control theory. Our proposed method makes the system capable of selecting the proper frequency and voltage settings in order to reduce the energy consumption, while guaranteeing hard real-time requirements in the presence of unpredictable workload fluctuations and faults. In the proposed method, the available slack-time is exploited by a feedback-based DVS at runtime to reduce the energy consumption. Furthermore, some slack-time is reserved for re-execution in case of faults. The simulation results show that compared with the traditional DVS method, our proposed method not only provides up to 59% energy saving, but also satisfies hard real-time constraints. Our proposed method is also effective in harnessing the static energy. The transition overheads are also taken into account in our simulation experiments.

CMDP based adaptive power management in server clusters

Many high-end computing systems use a large number of power-hungry commercial components to improve performance. Thus power reduction and energy conservation are important in these systems for the purpose of reducing operating costs. Two mechanisms are commonly applied to power reduction in these systems: Dynamic Voltage/Frequency Scaling (DV/FS) and server ON/OFF controlling: Vary-On Vary-Off (VOVF). Previous work addressed the power management issue as an optimization problem and utilized these two mechanisms either separately or together to reduce power consumption. In this paper, we construct a Constrained Markov Decision Process (CMDP) model and propose a CMDP based adaptive power management strategy in web server clusters. Our proposed strategy can greatly reduce the online computation time through an offline initialization process, the online adaptive server adjustment process will further improve system performance and deal with the dynamic changes of workload. It also combines merits from both DV/FS and VOVF mechanisms for better power reduction. The transition overhead caused by server allocation and frequency adjustment is taken into consideration in this work. The CMDP based adaptive power management strategy provides a controllable, predictable, and quantitative control over power consumption with theoretically guaranteed service performance. It is further evaluated via extensive simulations and justified by real workload trace data. The results illustrate the effectiveness and efficiency of our proposed work.

저전력 대기를 사용하는 OBS 망에서 에너지 절감을 위한 상태 천이 결정 알고리즘

Recently, many researchers have studied to solve the energy consumption of network equipments since the interest of Green IT has been increased. In this paper, we apply Low Power Idle (LPI) to OBS network to reduce energy consumption of network devices. Many previous researches have focused on maximizing the sleep time of network equipments to increase the energy saving efficiency of LPI. But transition overhead caused by LPI might not only depreciate the performance of energy saving but also increase packet delay. In this paper, Transition Decision (TD) algorithm is proposed to improve energy saving efficiency by reducing the number of unnecessary transition and guarantee the required QoS such as packet delay. To evaluate the performance of proposed algorithm, we model OBS edge router with LPI by OPNET and analyze the performance of the proposed algorithm in views of energy saving, transition count and average packet delay.

Developing resource consolidation frameworks for moldable virtual machines in clouds

This paper considers the scenario where multiple clusters of Virtual Machines (i.e., termed Virtual Clusters) are hosted in a Cloud system consisting of a cluster of physical nodes. Multiple Virtual Clusters (VCs) cohabit in the physical cluster, with each VC offering a particular type of service for the incoming requests. In this context, VM consolidation, which strives to use a minimal number of nodes to accommodate all VMs in the system, plays an important role in saving resource consumption. Most existing consolidation methods proposed in the literature regard VMs as “rigid” during consolidation, i.e., VMs’ resource capacities remain unchanged. In VC environments, QoS is usually delivered by a VC as a single entity. Therefore, there is no reason why VMs’ resource capacity cannot be adjusted as long as the whole VC is still able to maintain the desired QoS. Treating VMs as “moldable” during consolidation may be able to further consolidate VMs into an even fewer number of nodes. This paper investigates this issue and develops a Genetic Algorithm (GA) to consolidate moldable VMs. The GA is able to evolve an optimized system state, which represents the VM-to-node mapping and the resource capacity allocated to each VM. After the new system state is calculated by the GA, the Cloud will transit from the current system state to the new one. The transition time represents overhead and should be minimized. In this paper, a cost model is formalized to capture the transition overhead, and a reconfiguration algorithm is developed to transit the Cloud to the optimized system state with low transition overhead. Experiments have been conducted to evaluate the performance of the GA and the reconfiguration algorithm.

A Polynomial Based Approach to Wakeup Time and Energy Estimation in Power-Gated Logic Clusters

Power gating has emerged as a promising technique for reduction of leakage current in sub-100 nm CMOS circuits. To identify power gating opportunities within a system, logic circuits must be charac- terized for design overheads they present when power gating structures are inserted. In this paper, models are developed to quantify two mode transition overheads: wakeup time and wakeup energy using basic circuit parameters. First, a method to determine steady-state virtual-supply voltage in active mode is described and hence a model for virtual-supply voltage is presented based on certain heuristic approximations. Further, analytical expressions are derived for estimation of wakeup time and wakeup energy for a power-gated logic cluster using the proposed model. A key contribution of this paper is a study of nonlinear resistance based view of leakage current profiles of logic circuits. Finally, the application of proposed models to ISCAS85 benchmark circuits is demonstrated while also analyzing the accuracy of approximations used.

An Initial Evaluation of Energy Efficient Ethernet

In September 2010, the Energy Efficient Ethernet (IEEE 802.3az) standard was officially approved. This new standard introduces a low power mode for the most common Ethernet physical layer standards and is expected to provide large energy savings. In this letter, for the first time, Network Interface Cards (NICs) that implement Energy Efficient Ethernet (EEE) are used to measure energy savings with real traffic. The data presented will be useful to better estimate the energy savings that can be achieved when EEE is deployed. Existing analysis of EEE based on simulations predict a large overhead due to mode transitions between active and low power modes. The experimental results confirm that transition overheads can be significant, leading to almost full energy consumption even at low utilization levels. Therefore traffic patterns will play a key role in the energy savings achieved by EEE as it becomes deployed in the field.

Overhead-aware energy optimization for real-time streaming applications on multiprocessor System-on-Chip

In this article, we focus on solving the energy optimization problem for real-time streaming applications on multiprocessor System-on-Chip by combining task-level coarse-grained software pipelining with DVS (Dynamic Voltage Scaling) and DPM (Dynamic Power Management) considering transition overhead, inter-core communication and discrete voltage levels. We propose a two-phase approach to solve the problem. In the first phase, we propose a coarse-grained task parallelization algorithm called RDAG to transform a periodic dependent task graph into a set of independent tasks by exploiting the periodic feature of streaming applications. In the second phase, we propose a scheduling algorithm, GeneS, to optimize energy consumption. GeneS is a genetic algorithm that can search and find the best schedule within the solution space generated by gene evolution. We conduct experiments with a set of benchmarks from E3S and TGFF. The experimental results show that our approach can achieve a 24.4% reduction in energy consumption on average compared with the previous work.

Optimal Server Allocation and Frequency Modulation on Multi-Core Based Server Clusters

Multi-core processing has been widely used in high-performance computing systems. Power reduction and energy conservation are important in these systems to minimize operating cost. Two main strategies exist for power management: Dynamic Voltage/Frequency Scaling (DV/FS) and server number controlling: Vary-On Vary-Off (VOVF). In this paper, the authors explore the benefits of DV/FS and VOVF and apply them in a multi-core based server cluster. An optimization problem is formulated to get the optimal resource scheduling strategy given a parallel workload. The presented optimization model provides controllable and quantitative power consumption with theoretically guaranteed service performance. The authors further study the overhead of the proposed scheme and provide a Double Control Periods (DCP) method to compensate the transition overhead. The power optimization model is evaluated via extensive simulations. It is also justified by the real workload data trace. The results prove the effectiveness and efficiency of the proposed model.