Search

The growing demand for more computing resources has increased the overall energy consumption of computer systems. To support this increasing demand, power and energy consumption must be considered as a constraint on software execution. Modern architectures provide tools for managing the power constraints of a system directly. The Intel Power Cap is a relatively new tool developed to give users fine-grained control over power usage at the central processing unit (CPU) level. The complexity of these tools, in addition to the high variety of modern heterogeneous architectures, hinders predictions of the energy consumption and the performance of any target software. The application of power capping technologies usually leads to the bi-objective optimization problem for energy efficiency and execution time but optimal power constraints could also produce exceeding performance losses. Thus, methods and tools are needed to calculate the proper parameters for power capping technologies, and to optimize energy efficiency. We propose a methodology to analyze the performance and the energy efficiency trade-offs using this power cap technology for a given application. A Pareto front is extracted for the multi-objective performance and energy problem, which represents multiple feasible configurations for both objectives. An extensive experimentation is carried out to categorize the different applications to determine the overall optimal power cap configurations. We propose the use of machine learning (ML) clustering techniques to categorize each application in the target architecture. The use of ML allows us to automate the process and simplifies the effort required to solve the optimization problem. A practical case is presented where we categorize the applications using ML techniques, with the possibility of adding a new application into an existing categorization. [ABSTRACT FROM AUTHOR]

Currently,due to the wide application of multi-tenancy,containerization,virtualization and power over-subscription in data centers,the possibility of power attack is becoming increasingly higher.The main means of power attack is to run malicious codes to increase the power consumption of servers,storage device and network equipment to exceed the power limit of a distribution system.And it causes server failure or circuit breaker trip,or even the interruption of the power supply system of the data centers.In order to reduce the risk of power attack on data center,this paper proposes a power capping method of performance equivalence configuration(PCPEC).This method takes advantage of the difference of power consumption in different configurations of virtual machines to implement the equivalent replacement of virtual machine configuration.Experiment result shows that PCPEC can reduce the dynamic power consumption of the server by 22.2%~29.6%,and the performance of most virtual machines increases by 2.12% after the replacement of resource configuration,thus effectively reducing the impact of power attack on the data center.

High-performance computing (HPC), according to its name, is traditionally oriented toward performance, especially the execution time and scalability of the computations. However, due to the high cost and environmental issues, energy consumption has already become a very important factor that needs to be considered. The paper presents a survey of energy-aware scheduling methods used in a modern HPC environment, starting with the problem definition, tackling various goals set up for this challenge, including a bi-objective approach, power and energy constraints, and a pure energy solution, as well as metrics related to the subject. Then, considered types of HPC systems and related energy-saving mechanisms are described, from multicore-processors/graphical processing units (GPU) to more complex solutions, such as compute clusters supporting dynamic voltage and frequency scaling (DVFS), power capping, and other functionalities. The main section presents a collection of carefully selected algorithms, classified by the programming method, e.g., machine learning or fuzzy logic. Moreover, other surveys published on this subject are summarized and commented on, and finally, an overview of the current state-of-the-art with open problems and further research areas is presented. [ABSTRACT FROM AUTHOR]

Power and energy is the first-class design constraint for multi-core processors and is a limiting factor for future-generation supercomputers. While modern processor design provides a wide range of mechanisms for power and energy optimization, it remains unclear how software can make the best use of them. This article presents a novel approach for runtime power optimization on modern multi-core systems. Our policy combines power capping and uncore frequency scaling to match the hardware power profile to the dynamically changing program behavior at runtime. We achieve this by employing reinforcement learning (RL) to automatically explore the energy-performance optimization space from training programs, learning the subtle relationships between the hardware power profile, the program characteristics, power consumption and program running times. Our RL framework then uses the learned knowledge to adapt the chip's power budget and uncore frequency to match the changing program phases for any new, previously unseen program. We evaluate our approach on two computing clusters by applying our techniques to 11 parallel programs that were not seen by our RL framework at the training stage. Experimental results show that our approach can reduce the system-level energy consumption by 12 percent, on average, with less than 3 percent of slowdown on the application performance. By lowering the uncore frequency to leave more energy budget to allow the processor cores to run at a higher frequency, our approach can reduce the energy consumption by up to 17 percent while improving the application performance by 5 percent for specific workloads. [ABSTRACT FROM AUTHOR]

In dynamically adaptive streaming over HTTP (DASH), which is the de facto standard for streaming, each video is divided into segments, and each segment is further transcoded into multiple bitrate versions. This allows a client device to select the most appropriate bitrate version that matches the network bandwidth to avoid jitters or stalls. However, Wi-Fi download of a high-bitrate version may consume significant energy, especially when network conditions are good. To address this, we propose a new streaming method that limits the energy consumed by mobile devices but maintains an acceptable video quality. First, we derive a power model to analyze how bitrate selection affects power consumption in smartphones. Based on this, we propose two algorithms that determine the bitrate of each segment with the aim of maximizing overall video quality while limiting energy consumption. We use dynamic programming and heuristics to address the tradeoff between algorithm complexity and video quality. The proposed scheme was implemented on an Android-based DASH streaming platform, and various issues were resolved to cope with varying network conditions. Experimental results demonstrated that our scheme effectively optimized the video quality while limiting the energy consumption. For example: 1) our scheme uses 4% and 10% less power than DASH while maintaining an excellent video quality, and 2) the average difference between estimated and actual power consumption is 0.8%, thus keeping a precise energy bound.

High-performance computing (HPC), according to its name, is traditionally oriented toward performance, especially the execution time and scalability of the computations. However, due to the high cost and environmental issues, energy consumption has already become a very important factor that needs to be considered. The paper presents a survey of energy-aware scheduling methods used in a modern HPC environment, starting with the problem definition, tackling various goals set up for this challenge, including a bi-objective approach, power and energy constraints, and a pure energy solution, as well as metrics related to the subject. Then, considered types of HPC systems and related energy-saving mechanisms are described, from multicore-processors/graphical processing units (GPU) to more complex solutions, such as compute clusters supporting dynamic voltage and frequency scaling (DVFS), power capping, and other functionalities. The main section presents a collection of carefully selected algorithms, classified by the programming method, e.g., machine learning or fuzzy logic. Moreover, other surveys published on this subject are summarized and commented on, and finally, an overview of the current state-of-the-art with open problems and further research areas is presented.

Rapid growth of demand for remote computational power, along with high energy costs and infrastructure limits, has led to treating power usage as a primary constraint in data centers. Especially, recent challenges related to development of exascale systems or autonomous edge systems require tools that will limit power usage and energy consumption. This paper presents a power capping method that allows operators to quickly adjust the power usage to external conditions and, at the same time, to reduce energy consumption and negative impact on performance of applications. We propose an optimization model and both heuristic and exact methods to solve this problem. We present an evaluation of power capping approaches supported by results of application benchmarks and experiments performed on new heterogeneous servers. [ABSTRACT FROM AUTHOR]

Summary: Multithreaded applications facilitate the exploitation of the computing power of multicore architectures. On the other hand, these applications can become extremely energy‐intensive, in contrast with the need for limiting the energy usage of computing systems. In this article, we explore the design of techniques enabling multithreaded applications to maximize their performance under a power cap. We consider two control parameters: the number of cores used by the application, and the core power state. We target the design of an autotuning power‐capping technique with minimal intrusiveness and high portability, which is agnostic about the workload profile of the application. We investigate two different approaches for building the strategy for selecting the best configuration of the parameters under control, namely a heuristic approach and a model‐based approach. Through an extensive experimental study, we evaluate the effectiveness of the proposed technique considering two different selection strategies, and we compare them with existing solutions. [ABSTRACT FROM AUTHOR]

Power capping is an important solution to keep the system within a fixed power constraint. However, for the over-provisioned and power-constrained systems, especially the future exascale supercomputers, powercap needs to be reasonably allocated according to the workloads of compute nodes to achieve trade-offs among performance, energy and powercap. Thus it is necessary to model performance and energy and to predict the optimal powercap allocation strategies. Existing power allocation approaches have insufficient granularity within nodes. Modeling approaches usually model performance and energy separately, ignoring the correlation between objectives, and do not expose the Pareto-optimal powercap configurations. Therefore, this article combines the powercap with uncore frequency scaling and proposes an approach to predict the Pareto-optimal powercap configurations on the power-constrained system for input MPI and OpenMP parallel applications. Our approach first uses the elaborately designed micro-benchmarks and a small number of existing benchmarks to build the training set, and then applies a multi-objective machine learning algorithm which combines the stacked single-target method with extreme gradient boosting to build multi-objective models of performance and energy. The models can be used to predict the optimal processor and memory powercap settings, helping compute nodes perform fine-grained powercap allocation. When the optimal powercap configuration is determined, the uncore frequency scaling is used to further optimize the energy consumption. Compared with the reference powercap configuration, the predicted optimal configurations predicted by our method can achieve an average powercap reduction of 31.35 percent, an average energy reduction of 12.32 percent, and average performance degradation of only 2.43 percent. [ABSTRACT FROM AUTHOR]

Recent analyses show an increase in data center (DC) service downtime due to the increase in power supply outages caused by overloaded uninterruptable power supplies (UPSs). As a solution to the UPS overload problem, this letter presents a developed power distributor manager (PDM) that caps the power consumption of the dc by enforcing a power restriction on the running servers based on a UPS power minimization cost function. The PDM proposed controls the power consumption per UPS in a dc by means of processor time-sharing and dynamic frequency scaling, which eliminates UPS overloading problems, and thus reduces server outages. [ABSTRACT FROM AUTHOR]

Abstract An array of ten 1 MW turbines, with a one diameter inter-turbine spacing, is simulated in a hypothetical tidal channel using the depth-averaged flow solver TELEMAC-2D. It has been shown for idealised turbines that mean power extraction from tidal channels can be enhanced by tuning turbine resistance for a given channel. Blade element momentum theory is used to analyse turbine control strategies such as power capping, which can allow turbine operation to better match the dynamics of the underlying tidal resource. It is shown that choosing control strategies that limit array thrust helps sustain higher channel flow speeds, and hence can lead higher mean power production than other strategies. This has an impact on tidal farm size, where mean turbine power reduces as the number of turbines increases, due to the reducing flow rate in the channel. Adding turbines beyond the channel- and turbine-specific optimal chokes the flow, reducing overall farm power. To limit environmental impact it may be desirable to constrain the maximum flow-rate reduction. Under a 10% flow-rate reduction constraint, the lower thrust per turbine under pitch-to-feather control allows significantly more turbines to be installed and greater farm power than either pitch-to-stall or no power capping strategy. Highlights • Available tidal farm power is a function of farm thrust and channel dynamics • Effects of pitch-to-feather, pitch-to-stall, and no power capping are compared • Pitch-to-feather power capping achieves lowest thrust through tidal cycle • Environmental impact can be mitigated through choice of power control strategy [ABSTRACT FROM AUTHOR]

Abstract This paper considers the optimal strategy for the design and operation of tidal stream turbine farms. Using an idealised model we explore the implications of limiting the peak power and peak thrust of the turbines. We consider the impact on power production and on the environmental impact. Our results suggest that a combined power and thrust-capping strategy should be adopted in the design of tidal stream turbine farms. Highlights • This work looks at extending the existing literature on power capping of tidal stream turbines to look at thrust capping. • The influence of power capping, thrust capping, and the two combined, are analysed for an idealised tidal channel. • This work show that sizing tidal devices based on the combined capping would be more economic. • The results show that combined capping can reduce the requirements in device capacity and support structure. • The combined capping can also help to reduce the change to the marine environment. [ABSTRACT FROM AUTHOR]

Constructive interference between tidal stream turbines in multi-rotor fence configurations arrayed normally to the flow has been shown analytically, computationally, and experimentally to enhance turbine performance. The increased resistance to bypass flow due to the presence of neighbouring turbines allows a static pressure difference to develop in the channel and entrains a greater flow rate through the rotor swept area. Exploiting the potential improvement in turbine performance requires that turbines either be operated at higher tip speed ratios or that turbines are redesigned in order to increase thrust. Recent studies have demonstrated that multi-scale flow dynamics, in which a distinction is made between device-scale and fence-scale flow events, have an important role in the physics of flow past tidal turbine fences partially spanning larger channels. Although the reduction in flow rate through the fence as the turbine thrust level increases has been previously demonstrated, the within-fence variation in turbine performance, and the consequences for overall farm performance, is less well understood. The impact of turbine design and operating conditions, on the performance of a multi-rotor tidal fence is investigated using Reynolds-Averaged Navier-Stokes embedded blade element actuator disk simulations. Fences consisting of four, six, and eight turbines are simulated, and it is demonstrated that the combination of device- and fence-scale flow effects gives rise to cross-fence thrust and power variation. These cross-fence variations are also a function of turbine thrust, and hence design conditions, although it is shown simple turbine control strategies can be adopted in order to reduce the cross-fence variations and improve overall fence performance. As the number of turbines in the fence, and hence fence length, increases, it is shown that the turbines may be designed or operated to achieve higher thrust levels than if the turbines were not deployed in a fence configuration. [ABSTRACT FROM AUTHOR]

As the scale and density of data centers continue to grow, cost-effective data center management (DCM) is becoming a significant challenge for enterprises hosting large-scale online and cloud services.Machines need to be monitored, and the scale of operations mandates an automated management with high reliability and real-time performance. The limitations of today's typical DCM network are many-fold. Primarily, it is a fixed wired network, and hence scaling it for a large number of servers increases its cost. In addition, with server densities increasing over recent years, this network also has to be cabled correctly and themanagement of this network parallels the complexity of managing a data network, since it needs to be networked with multiple switches and routers. In this article, we propose a wireless sensor network as a cost-effective networking solution for DCM while satisfying the reliability and latency performance requirements of DCM. We have developed CapNet, a real-time wireless sensor network for power capping, a time-critical DCM function for power management in a cluster of servers. CapNet employs an efficient event-driven protocol that triggers data collection only on the detection of a potential power capping event. We deploy and evaluate CapNet in a data center. Using server power traces, our experimental results on a cluster of 480 servers inside the data center show that CapNet can meet the real-time requirements of power capping. CapNet demonstrates the feasibility and efficacy of wireless sensor networks for time-critical DCM applications. [ABSTRACT FROM AUTHOR]

Abstract A key hydrodynamic difference between tidal current and wind turbines is the volume-flux constrained flow field in which tidal turbines operate and the resulting streamwise static pressure difference that develops in the flow passage. Blade Element Momentum (BEM) theory is extended to account analytically for the effects of blockage and the development of the static pressure difference in the flow passage and shows agreement in thrust and power predictions to within ± 3 % of equivalent blade resolved simulations. The confined flow BEM model is employed to study two different power capping strategies: varying the rotational speed with fixed pitch blades; and pitching the blades to feather at constant rotational speed. Pitch-to-feather achieves reduced thrust above rated flow speed which leads to a greater extractable resource than achievable with overspeed control, due to the feedback between device thrust and available tidal resource. Flow confinement is shown to reduce the flow speed at which rated power occurs, and increases the rotor loads and power below rated conditions. It is also shown that root bending moments, which affect fatigue damage rates, increase with flow confinement. [ABSTRACT FROM AUTHOR]

Abstract A fully blade resolved CFD model with a RANS formulation using the k − ω SST turbulence model was used to study the tidal turbine performance in the presence of two support structures with cylindrical and elliptical cross-sections. The integrated rotor force coefficients were higher in the presence of the cylindrical support structure than the elliptical support due to the higher opposing thrust from the cylinder in the channel increasing the local flow velocity near the top half of the rotor. The angle of attack on the blade decreases as it approaches the support structure due to the reduction in stream-wise flow velocity just ahead of the support structure. This drop is higher for cylindrical support than the elliptical support due to its larger opposing thrust to the flow causing a larger drop in stream-wise velocity. The drop in the angle of attack due to the presence of support structures causes a significant drop in the blade sectional forces as it approaches the support structure. The presence of rotor also causes a drop in the forces on the support structure. The mean stream-wise force coefficients of the cylindrical and elliptical support structures are 0.21 and 0.12 respectively, which are much lower than if they were in open water without a rotor. In case of both the support structures, the stream-wise sectional forces behind the rotor swept area are lower than that in open-water whereas the mean cross-stream forces are higher. The thrust from the rotor and the support structure has a direct influence on the wake velocity. The velocity deficit just in-front of the support structure is 20% higher for the cylindrical support than the elliptical support due to its higher opposing thrust to the flow, and further downstream, the case with cylindrical support has higher velocity deficit followed by elliptical support and then the case with no support structure. Highlights • Implementation of fully blade resolved RANS formulation with k − ω SST turbulence model to study the interaction effects between turbine rotor and two different support structure shapes. • Integrated rotor force coefficients were higher in the presence of the cylindrical support structure than the elliptical support structure. • Presence of rotor causes a drop in the mean stream-wise force coefficients of the support structures. • The velocity deficit just in-front of the support structure is 20% higher for the cylindrical support than the elliptical support. [ABSTRACT FROM AUTHOR]

Abstract Power capping are increasingly being deployed in modern processors to meet performance/power requirements of new workloads. These computing systems are limited in their power dissipation, demanding a dependable power management scheme to guarantee the system's efficiency and dependability. Although several ad-hoc and heuristic power management approaches can be found in the literature, their main shortcoming is the lack of formal guarantees to ensure dependability of the processors. Control-theoretic approaches promise flexibility and robustness for DVFS strategies. However, the creation of a responsive yet stable controller requires the often neglected tasks of proper system identification and performance analysis for target applications. This paper presents dependability evaluation of Single-Input Single-Output (SISO) controllers for power capping on single-core as well as multi-core processor architectures. We evaluate responsiveness of different class of applications to computer system control inputs (i.e., DVFS). We illustrate the feasibility of SISO controllers for power capping using the Sniper simulator running PARSEC, SPLASH2 and a set of custom microbenchmarks. Based on our observations, we provide guidelines for developing stable and robust SISO controllers for power capping, show the scenarios where simple classic SISO controllers might not be effective, and identify early symptoms that may result in instability for power capping controllers. [ABSTRACT FROM AUTHOR]

An embedded Reynolds-Averaged Navier-Stokes blade element actuator disk model is used to investigate the hydrodynamic design of tidal turbines and their performance in a closely spaced cross-stream fence. Turbines designed for confined flows are found to require a larger blade solidity ratio than current turbine design practices imply in order to maximise power. Generally, maximum power can be increased by operating turbines in more confined flows than they were designed for, although this also requires the turbines to operate at a higher rotational speed, which may increase the likelihood of cavitation inception. In-array turbine performance differs from that predicted from single turbine analyses, with cross-fence variation in power and thrust developing between the inboard and outboard turbines. As turbine thrust increases the cross-fence variation increases, as the interference effects between adjacent turbines strengthen as turbine thrust increases, but it is observed that cross-stream variation can be mitigated through strategies such as pitch-to-feather power control. It was found that overall fence performance was maximised by using turbines designed for moderately constrained (blocked) flows, with greater blockage than that based solely on fence geometry, but lower blockage than that based solely on the turbine and local flow passage geometry to balance the multi-scale flow phenomena around tidal fences.

Recent studies have proposed to dynamically reshape the power demand curve of a data center (i.e., power shaving) with energy storage devices, particularly uninterruptible power supply (UPS) batteries. Power shaving can be used to limit the peak power demand in a data center, in order to reduce both the power infrastructure investment (i.e., cap-ex) and the electricity bills (i.e., op-ex). However, power shaving requires the UPS batteries to be frequently charged/discharged, which is known to compromise the battery lifetime and availability. This paper presents a detailed quantitative study that explores different options to integrate supercapacitor (SC) with batteries for cost-efficient energy storage. Compared with batteries, SC allows more charge/discharge cycles and has a higher power density, which are desirable for fast power shaving. However, SC also has undesirable characteristics (e.g., relatively high self-discharging rate and cost). Therefore, we quantitatively compare three possible energy storage options (i.e., Battery-only, SC-only, and Battery+SC ) in detail, with different SC self-discharging rate assumptions. SC options (SC-only and Battery+SC) are shown to be more cost-efficient designs, saving the energy storage cost by 34 percent, on average, compared with Battery-only. For a 10 MW data center in a 10-year period, the savings can be converted to $3 M in total cost of ownership (TCO) reduction by allowing more servers to be deployed. In addition, we also propose the integration of energy storage with dynamic voltage and frequency scaling (DVFS) to cap the peak power demand (i.e., power capping). Specifically, we comparatively study four power capping algorithms and discuss their applicable scenarios. Finally, we introduce our proof-of-concept SC physical testbed and present preliminary hardware testing results. [ABSTRACT FROM PUBLISHER]

Shared resource contention has been a major performance issue for CMPs. In this paper, we tackle the power contention problem in power constrained CMPs by considering and treating power as a first-class shared resource. Power contention occurs when multiple processes compete for power, and leads to degraded system performance. In order to solve this problem, we develop a shared resource contention-aware scheduling algorithm that mitigates the contention for power and the shared memory subsystem at the same time. The proposed scheduler improves system performance by balancing the shared resource usage among scheduling groups. Evaluation results across a variety of multiprogrammed workloads show performance improvements over a state-of-the-art scheduling policy which only considers memory subsystem contention. [ABSTRACT FROM PUBLISHER]

We present a two-level implementation of an infrastructure that allows performance maximization under a power-cap on multi-application environments with minimal user intervention. At the application level, we integrate bar (Power Budget-Aware Runtime Scheduler) into existing task-based runtimes, e.g. OpenMP; bar implements combined software/hardware techniques (thread malleability and DVFS) to maximize the application performance without violating a granted power budget. At a higher level, we introduce barman (Power Budget-Aware Resource Manager), a system-wide software able to manage resources globally, gathering power needs of registered applications, and redistributing the available overall power budget across them. The combination and co-operative operation of both pieces of software yields performance and energy efficiency improvements on environments in which power capping is established globally, and also granted asymmetrically to different co-existing applications. This behaviour is demonstrated to be stable under different workloads (a selection of task-based scientific applications and PARSEC benchmarks are tested) and different levels of power capping. • We show bar , a task-based runtime designed to maximize performance under a power cap • bar is able to dynamically adjust both software and hardware mechanisms simultaneously • We introduce barman as a power budget-aware co-scheduling mechanism • barman re-distributes the available power budget across applications without user intervention. [ABSTRACT FROM AUTHOR]

The power supplied to machine rooms tends to be over-provisioned because it is specified in practice not by workload demands but rather by high energy LINPACK runs or nameplate power estimates. This results in a considerable amount of trapped power capacity —excess power infrastructure. Instead of being wasted, this trapped power capacity should be reclaimed to accommodate more compute nodes in the machine room and thereby increase system throughput. But to do this we need the ability to enforce a system-wide power cap. In this paper, we present TracSim , a full-system simulator that enables users to measure trapped power capacity and evaluate the performance of different policies for scheduling parallel tasks under a power cap. TracSim simulates the execution environment of a production HPC cluster at Los Alamos National Laboratory (LANL). TracSim enables users to specify the system topology, hardware configuration, power cap, and task workload and to develop resource configuration and task scheduling policies aimed at maximizing machine-room throughput while keeping power consumption under a power cap by exploiting CPU throttling techniques. We use real measurements from the LANL cluster to set TracSim ’s configuration parameters. We leverage TracSim to implement and evaluate four resource scheduling policies. Simulation results indicate the performance of those policies and quantify the amount of trapped capacity that can effectively be reclaimed. [ABSTRACT FROM AUTHOR]

Aggressive technology scaling triggers novel challenges to the design of multi-/many-core systems, such as limited power budget and increased reliability issues. Today’s many-core systems employ dynamic power management and runtime mapping strategies trying to offer optimal performance while fulfilling power constraints. On the other hand, due to the reliability challenges, online testing techniques are becoming a necessity in current and near future technologies. However, state-of-the-art techniques are not aware of the other power/performance requirements. This paper proposes a power-aware non-intrusive online testing approach for many-core systems. The approach schedules software based self-test routines on the various cores during their idle periods, while honoring the power budget and limiting delays in the workload execution. A test criticality metric, based on a device aging model, is used to select cores to be tested at a time. Moreover, power and reliability issues related to the testing at different voltage and frequency levels are also handled. Extensive experimental results reveal that the proposed approach can i) efficiently test the cores within the available power budget causing a negligible performance penalty, ii) adapt the test frequency to the current cores’ aging status, and iii) cover available voltage and frequency levels during the testing. [ABSTRACT FROM AUTHOR]

Power shaving has recently been proposed to dynamically shave the power peaks of a data center with energy storage devices (ESD), such that more servers can be safely hosted. In addition to the reduction of capital investment (cap-ex), power shaving also helps cut the electricity bills (op-ex) of a data center by reducing the high utility tariffs related to peak power. However, existing work on power shaving focuses exclusively on electrical ESDs (e.g., UPS batteries) to shave the server-side power demand. In this paper, we propose TE-Shave, a generalized power shaving framework that exploits both UPS batteries and a new knob, thermal energy storage (TES) tanks equipped in many data centers. Specifically, TE-Shave utilizes stored cold water or ice to manipulate the cooling power, which accounts for 30-40 percent of the total power cost of a data center. Our extensive evaluation with real-world workload traces shows that TE-Shave saves cap-ex and op-ex up to $2,668/day and $825/day, respectively, for a data center with 17,920 servers. Even for future data centers that are projected to have more efficient cooling and thus a smaller portion of cooling power, e.g., a quarter of today’s level, TE-Shave still leads to 28 percent more savings than existing work that focuses only on the server-side power. TE-Shave is also coordinated with traditional TES solutions for further reduced op-ex, and integrated with processor throttling to cap the power draw (i.e., power capping). Our hardware testbed results show that TE-Shave can improve the system performance up to 23 percent. [ABSTRACT FROM PUBLISHER]

As high performance computing (HPC) continues to grow in scale and complexity, energy becomes a critical constraint in the race to exascale computing. The days of “performance at all cost” are coming to an end. While performance is still a major objective, future HPC will have to deliver desired performance under the energy constraint. Among various power management methods, power capping is a widely used approach. Unfortunately, the impact of power capping on system performance, user jobs, and power-performance efficiency are not well studied due to many interfering factors imposed by system workload and configurations. To fully understand power management in extreme scale systems with a fixed power budget, we introduce a power-performance modeling tool named PuPPET (Power Performance PETri net) . Unlike the traditional performance modeling approaches such as analytical methods or trace-based simulators, we explore a new approach–colored Petri nets–for the design of PuPPET. PuPPET is fast and extensible for navigating through different configurations. More importantly, it can scale to hundreds of thousands of processor cores and at the same time provide high levels of modeling accuracy. We validate PuPPET by using system traces (i.e., workload log and power data) collected from the production 48-rack IBM Blue Gene/Q supercomputer at Argonne National Laboratory. Our trace-based validation demonstrates that PuPPET is capable of modeling the dynamic execution of parallel jobs on the machine by providing an accurate approximation of energy consumption. In addition, we present two case studies of using PuPPET to study power-performance tradeoffs on petascale systems. [ABSTRACT FROM AUTHOR]

To achieve the exaFLOPS performance within a contained power budget, next generation supercomputers will feature hundreds of millions of components operating at low- and near-threshold voltage. As the probability that at least one of these components fails during the execution of an application approaches certainty, it seems unrealistic to expect that any run of a scientific application will not experience some performance faults. We believe that there is need of a new generation of light-weight performance and debugging tools that can be used online even during production runs of parallel applications and that can identify performance anomalies during the application execution. In this work we propose the design and implementation of a monitoring system that continuously inspects the evolution of running applications and the health of the system. To achieve minimum runtime overhead while maintaining the desired level of flexibility, we propose a decoupled approach in which accurate monitoring is performed at kernel-level while performance anomaly disambiguation and corrective actions are performed at user-level. We evaluate our monitoring system on a 128-core AMD Interlagos system: First, we show that the runtime overhead of the monitoring system is negligible (0-2%). Then we show how our system can be used to precisely identify performance faults in three different scenarios: OS noise, application co-scheduling and dynamic power capping. [ABSTRACT FROM AUTHOR]

As supercomputers continue to grow in size and power consumption, the ability to understand application run time performance and energy/power usage is of growing importance. In the future it may be necessary for systems to operate under power caps imposed by facilities due to external influences such as renewable power generation (e.g. solar) impacting energy availability at an affordable cost during certain times of day and more frequent natural disasters due to climate change causing power transmission disruptions. However, current energy consumption and power characterization metrics like energy-delay products do not express all of the characteristics of an application that are relevant to understanding the impact of power capping on performance. In this study, we (1) characterize the useful features of a metric that effectively captures time-to-solution and power performance dynamics; and (2) design and evaluate a novel set of application power efficiency (APE) metrics that accounts for time-to-completion, power utilization and power variability. Power utilization quantifies a system's trapped capacity (unused power budget) and power variability quantifies an application's variation of power consumption over time. We demonstrate how APE can be used to quantify application characteristics which, in turn, enables reasoning about the impacts of power capping on an application. • A method for understanding application performance in power-constrained environments. • APE metrics and an analysis of the strengths and weaknesses of existing metrics. • A demonstration that APE quantifies important application power performance features. [ABSTRACT FROM AUTHOR]