Power-aware predictive models of hybrid (MPI/OpenMP) scientific applications on multicore systems

Lively, Charles; Wu, Xingfu; Taylor, Valerie; Moore, Shirley; Chang, Hung-Ching; Su, Chun-Yi; Cameron, Kirk W.

doi:10.1007/s00450-011-0190-0

Cited by 54 publications

(49 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Lively et al [17] employ 40 PMCs in their predictive model. They use an elaborate statistical methodology to select PMCs.…”

Section: Techniques For Selection Of Pmcs For Energy Predictive Modelingmentioning

confidence: 99%

“…It should be mentioned that some of these non-additive PMCs such as P AP I L1 ICM and P AP I L2 ICM have been widely used in energy and performance predictive models [2,4,7,17,27,28]. These represent L1 and L2 instruction cache misses.…”

Section: Uops Executed Port Port 7 Core Uops Executed Port Data Portsmentioning

confidence: 99%

“…On this server, the PAPI tool [19] provides 53 hardware performance events. The Likwid tool [17,23] provides 167 PMCs. This includes events for uncore and micro-operations (µops) of CPU cores specific to Haswell architecture that are not provided by PAPI.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Additivity: A Selection Criterion for Performance Events for Reliable Energy Predictive Modeling

Shahid

Fahad

Reddy

et al. 2017

JSFI

View full text Add to dashboard Cite

Performance events or performance monitoring counters (PMCs) are now the dominant predictor variables for modeling energy consumption. Modern hardware processors provide a large set of PMCs. Determination of the best subset of PMCs for energy predictive modeling is a nontrivial task given the fact that all the PMCs can not be determined using a single application run. Several techniques have been devised to address this challenge. While some techniques are based on a statistical methodology, some use expert advice to pick a subset (that may not necessarily be obtained in one application run) that, in experts' opinion, are significant contributors to energy consumption. However, the existing techniques have not considered a fundamental property of predictor variables that should have been applied in the first place to remove PMCs unfit for modeling energy. We address this oversight in this paper.We propose a novel selection criterion for PMCs called additivity, which can be used to determine the subset of PMCs that can potentially be used for reliable energy predictive modeling. It is based on the experimental observation that the energy consumption of a serial execution of two applications is the sum of energy consumptions observed for the individual execution of each application. A linear predictive energy model is consistent if and only if its predictor variables are additive in the sense that the vector of predictor variables for a serial execution of two applications is the sum of vectors for the individual execution of each application. The criterion, therefore, is based on a simple and intuitive rule that the value of a PMC for a serial execution of two applications is equal to the sum of its values obtained for the individual execution of each application. The PMC is branded as non-additive on a platform if there exists an application for which the calculated value differs significantly from the value observed for the application execution on the platform. The use of non-additive PMCs in a model renders it inconsistent.We study the additivity of PMCs offered by the popular state-of-the-art tools, Likwid and PAPI, by employing a detailed experimental methodology on a modern Intel Haswell multicore server CPU. We show that many PMCs in Likwid and PAPI that are widely used in models as key predictor variables are non-additive. This brings into question the reliability and the reported prediction accuracy of these models.

show abstract

“…Lively et al [17] employ 40 PMCs in their predictive model. They use an elaborate statistical methodology to select PMCs.…”

Section: Techniques For Selection Of Pmcs For Energy Predictive Modelingmentioning

confidence: 99%

Section: Uops Executed Port Port 7 Core Uops Executed Port Data Portsmentioning

confidence: 99%

See 1 more Smart Citation

Additivity: A Selection Criterion for Performance Events for Reliable Energy Predictive Modeling

Shahid

Fahad

Reddy

et al. 2017

JSFI

View full text Add to dashboard Cite

show abstract

“…A number of projects use power/energy models based on hardware counter data [6], [7], [8] to determine optimal frequency settings for an application. Green Queue relies on application characteristics and power models based on those characteristics to make frequency scaling decisions.…”

Section: Related Workmentioning

confidence: 99%

Green Queue: Customized Large-Scale Clock Frequency Scaling

Tiwari

Laurenzano

Peraza

et al. 2012

2012 Second International Conference on Cloud and Green Computing

View full text Add to dashboard Cite

Abstract-We examine the scalability of a set of techniques related to Dynamic Voltage-Frequency Scaling (DVFS) on HPC systems to reduce the energy consumption of scientific applications through an application-aware analysis and runtime framework, Green Queue. Green Queue supports making CPU clock frequency changes in response to intra-node and internode observations about application behavior. Our intra-node approach reduces CPU clock frequencies and therefore power consumption while CPUs lacks computational work due to inefficient data movement. Our inter-node approach reduces clock frequencies for MPI ranks that lack computational work.We investigate these techniques on a set of large scientific applications on 1024 cores of Gordon, an Intel Sandybridgebased supercomputer at the San Diego Supercomputer Center. Our optimal intra-node technique showed an average measured energy savings of 10.6% and a maximum of 21.0% over regular application runs. Our optimal inter-node technique showed an average 17.4% and a maximum of 31.7% energy savings.

show abstract

“…Some applications are computationally intensive, other are data intensive, while other are hybrid of both. Energy-aware programming models, regarding various types of workloads and architectures, are required to develop energy efficient applications [21,22]. Programming approaches, based on optimal utilization of shared hardware on-chip resources, e.g.…”

mentioning

confidence: 99%

Energy Efficiency of Parallel Multicore Programs

Davidović¹,

Depolli²,

Lipić³

et al. 2016

SCPE

View full text Add to dashboard Cite

Abstract. The increasing energy consumption of large-scale high performance resources raises technical and economical concerns. A reduction of consumed energy in multicore systems is possible to some extent with an optimized usage of computing and memory resources that is tailored to specific HPC applications. The essential step towards more sustainable consumption of energy is its reliable measurements for each component of the system and selection of optimally configured resources for specific applications. This paper briefly surveys the current approaches for measuring and profiling power consumption in large scale systems. Then, a practical case study of a real-time power measurement of multicore computing system is presented on two real HPC applications: maximum clique algorithm and numerical weather prediction model. We assume that the computing resources are allocated in a HPC cloud on a pay-per-use basis. The measurements demonstrate that the minimal energy is consumed when all available cores (up to the scaling limit for a particular application) are used on their maximal frequencies and with threads binded to the cores.Key words: Energy efficiency, high-performance, maximum clique, numerical weather prediction AMS subject classifications. 65Y05, 68U201. Introduction. Nowadays, a large number of hard problems is being solved efficiently on large-scale computing systems such as clusters, grids and clouds. The main building blocks of such systems are highperformance computing (HPC) clusters, since they are furthermore embedded both into grid computing systems as well as cloud computing systems [1]. A great advantage of Graphical Processor Unit (GPU) accelerators over multicore platforms was demonstrated in terms of energy efficiency [2] with an overall energy consumption for an order of magnitude smaller for the GPUs. Inside the clusters, the many-core architectures, such as GPUs [3], as well as even more specialized data-flow computers [4] are getting an increasingly large supporting role but the main processing is still performed by the multi-core main processors because of simple programming model.In this research, we analyze the energy consumption profile of the processors on a two processor multi-core computer, while varying its workload.With the increasing number of components in large-scale computing systems the energy consumption is steeply increasing [5]. In computer systems, energy (measured in Joules) is represented as the electricity resource that can power the hardware components to do computation. On the other hand, power is the amount of energy consumed per unit time (measured in Joules per second or Watts). The ever-increasing energy consumption of large-scale computing systems causes higher operation costs (e.g. electricity bills) and has negative environmental impacts (e.g. carbon dioxide emissions). Therefore, when designing large-scale computing systems, the focus is often shifting from performance improvements to energy efficiency.Energy can be reduced at several levels of the distrib...

show abstract

Power-aware predictive models of hybrid (MPI/OpenMP) scientific applications on multicore systems

Cited by 54 publications

References 7 publications

Additivity: A Selection Criterion for Performance Events for Reliable Energy Predictive Modeling

Additivity: A Selection Criterion for Performance Events for Reliable Energy Predictive Modeling

Green Queue: Customized Large-Scale Clock Frequency Scaling

Energy Efficiency of Parallel Multicore Programs

Contact Info

Product

Resources

About