计算机科学 ›› 2024, Vol. 51 ›› Issue (9): 59-70.doi: 10.11896/jsjkx.231100015
李泽锴, 钟佳卿, 冯绍骏, 陈娟, 邓荣宇, 徐涛, 谭政源, 周柯杏, 朱鹏志, 马兆阳
LI Zekai, ZHONG Jiaqing, FENG Shaojun, CHEN Juan, DENG Rongyu, XU Tao, TAN Zhengyuan, ZHOU Kexing, ZHU Pengzhi, MA Zhaoyang
摘要： 建立高精度、低开销的CPU功耗模型对于计算机系统的功耗管理与功耗优化至关重要。一般认为训练集规模越大,CPU功耗模型精度越高。但有研究发现增大训练集规模不一定会提高功耗建模精度,有时甚至会导致精度下降,因此,如何选择功耗模型训练集以保证CPU功耗模型精度达到要求具有重要意义。文中提出一种基于聚类的训练集选择优化算法来解决上述问题,在有效保证CPU功耗建模精度的同时降低了CPU功耗建模的开销。该算法首先通过主成分分析将基于PMC的程序特征转换为p维向量特征空间,然后根据找到的最优聚类数按照程序特征对程序进行聚类,从每个聚类簇中选出代表程序;最后根据“单聚类簇内代表性最强原则”与“多聚类簇间代表程序数最少原则”形成最优训练集,模型精度相比Baseline精度有明显提高。在x86和ARM两类处理器平台上分别采用线性功耗建模和神经网络功耗建模两种方式,对算法进行了实验评估,实验结果表明所提算法的功耗建模精度有效显著提升。
中图分类号:
| [1]CHEN J,ZHOU W H,DONG Y,et al.Analyzing time-dimen-sion communication characterizations for representative scienti-fic applications on super computer systems[J].Frontiers of Computer Science,2019,13(6):1228-1242. [2]CHEN J,QI X,WU F,et al.More Bang for Your Buck:Boosting Performance with Capped Power Consumption[J].Tsinghua Science and Technology,2021,26(3):370-383. [3]WANG Z,TANG Y,CHEN J,et al.Energy wall for exascalesupercomputing[J].Computing and Informatics,2016,35(4):941-962. [4]TIWARI A,LAURENZANO M A,CARRINGTON L,et al.Modeling Power and Energy Usage of HPC Kernels[C]//IEEE International Parallel & Distributed Processing Symposium Workshops & Phd Forum.IEEE,2012. [5]ZHOU Z,ABAWAJY J H,LI F,et al.Fine-Grained EnergyConsumption Model of Servers Based on Task Characteristics in Cloud Data Center[J].IEEE Access,2018,6:27080-27090. [6]VON KISTOWSKI J,GROHMANN J,SCHMITT N,et al.Predicting server power consumption from standard rating results[C]//International Conference on Performance Engineering.2019. [7]LIN W,WU G,WANG X,et al.An artificial neural network approach to power consumption model construction for servers in cloud data centers[J].IEEE Transactions on Sustainable Computing,2019,5(3):329-340. [8]LI Y,HU H,WEN Y,et al.Learning-based power prediction for data centre operations via deep neural networks[C]//the 5th International Workshop.ACM,2016. [9]ZHANG Y,DONG Y,CHEN J,et al.Pmc-based dynamic adaptive cpu and dram power modeling[C]//International Confe-rence on Algorithm and Architectures for Parrallel Processing.2020:92-111. [10]SPEC.Standard performance evaluation corporation[EB/OL].2017.http://www.spec.org/cpu2017/. [11]LUSZCZEK P R,BAILEY D H,DONGARRA J J,et al.S12-the hpc challenge(hpcc) benchmark suite[Z].2006. [12]PARSEC.The princeton application repository for shared- me-mory computers[EB/OL].https://parsec.oden.utexas.edu/. [13]GRAPH500.Graph500[EB/OL].https://graph500.org/. [14]BRIAN C.The smg2000 benchmark code[EB/OL].https://asc.llnl.gov/computing_resources/purple/archive/benchmarks/smg/. [15]HPCG.High performance conjugate gradients[EB/OL].ht-tps://www.hpcg-benchmark.org/. [16]JACK D.Hpl-ai mixed-precision benchmark[EB/OL].https://icl.bitbucket.io/hpl-ai/. [17]ERIKSSON L,JOHANSSON E,WOLD S.On the selection of the training set in environmental qsar analysis when compounds are clustered[J].Journal of Chemometrics,2000,14:599-616. [18]GUTIERREZ M,TAMIR D,QASEM A.Evaluating NeuralNetwork Methods for PMC-based CPU Power Prediction[C]//International Multi-Conference on Computing in the Global Information Technology.2015. [19]MARBACH M,ONDUSKO R,RAMACHANDRAN R P,et al.Neural network classifiers and Principal Component Analysis for blind signal to noise ratio estimation of speech signals[C]//2009 IEEE International Symposium on Circuits and Systems(ISCAS).2009:97-100. [20]CALIN＇SKI T,HARABASZ J.A dendrite method for clusteranalysis[J].Communications in Statistics-theory and Methods,1974,3:1-27. [21]ZHANG F L,ZHOU H C,ZHANG J J,et al.Protocol classification algorithm based on improved cohesive hierarchical clustering [J].Computer Engineering and Science,2017,39(4):796-803. [22]XIE Z Y,XU X Q,WALKER M,et al.Apollo:An automated power modeling framework for run-time power introspection in high-volume commercial microprocessors[C]//54th Annual IEEE/ACM International Symposium on Microarchitecture.2021:1-14. [23]NIKOV K,MARTINEZ M,CHAMSKI Z,et al.Robust and accurate fine-grain power models for embedded systems with no on-chip pmu[J].arXiv.2106.00565,2021. [24]TIWARI A,LAURENZANO M A,CARRINGTON L,et al.Modeling Power and Energy Usage of HPC Kernels[C]//IEEE International Parallel & Distributed Processing Symposium Workshops & Phd Forum.IEEE,2012. [25]FENG X,GE R,CAMERON K W.Power and Energy Profiling of Scientific Applications on Distributed Systems[C]//19th IEEE International Parallel and Distributed Processing Sympo-sium.2005. [26]CONTRERAS G,MARTONOSI M.Power prediction for intel xscale processors using performance monitoring unit events[C]//Proceedings of the 2005 International Symposium on Low Power Electronics and Design.2005:221-226. [27]WU X,TAYLOR V.Utilizing Hardware Performance Counters to Model and Optimize the Energy and Performance of Large Scale Scientific Applications on Power-Aware Supercomputers[C]//IEEE International Parallel & Distributed Processing Symposium Workshops.IEEE,2016. [28]BIRCHER W L,VALLURI M,LAW J,et al.Runtime identifi-cation ofmicroprocessor energy saving opportunities[C]//International Symposium on Low Power Electronics & Design.IEEE,2005. [29]QU G,KAWABE N,USAMI K,et al.Function-level power estimation methodology for microprocessors[C]//Proceedings of the 37th Annual Design Automation Conference.2000:810-813. [30]LAROS J H,POKORNY P,DEBONIS D,et al.Powerinsight－a commodity power measurement capability[C]//2013 International Green Computing Conference Proceedings(IGCC).2013:1-6. [31]BEDARD D,LIM M Y,FOWLER R,et al.Powermon:Fine-grained and integrated power monitoring for commodity compu-ter systems[C]//IEEE Southeastcon.2010:479-484. [32]ZHENG X,JOHN L K,GERSTLAUER A,et al.Lacross:Learning-based analytical cross-platform performance and power prediction[J].International Journal of Parallel Programming,2017,45(6):1488-1514. [33]JOSEPH R,MARTONOSI M.Run-time power estimation inhigh performance microprocessors[C]//International Sympo-sium on Low Power Electronics Design.2001. |
|
||
首页