Performance and power modeling and prediction using MuMMI and 10 machine learning methods

Performance and power modeling and prediction using MuMMI and 10 machine learning methods
复制标题

使用 MuMMI 和 10 种机器学习方法进行性能和功耗建模和预测

DOI:
10.1002/cpe.7254
复制
发表时间:
2022
期刊:
Concurrency and Computation: Practice and Experience
影响因子:
--
通讯作者:
Lan, Zhiling
Lan, Zhiling
中科院分区:
--
文献类型:
--
作者:
Wu, Xingfu;Taylor, Valerie;Lan, Zhiling

文献摘要

参考文献

相似文献

高能效科学应用需要深入了解高性能计算系统功能如何影响应用的功率和性能。这种洞察力可能来自性能和功耗模型的开发。在本文中,我们使用建模和预测工具MuMMI(多线程建模基础设施)和10种机器学习方法来建模和预测性能和功耗,并比较它们的预测错误率。我们使用基于算法的容错线性代数代码和多级检查点容错热分布代码对阿贡国家实验室的Cray XC 40 Theta和IBM BG/Q Mira以及桑迪亚国家实验室的英特尔Haswell集群谢泼德进行建模和预测研究。我们的实验结果表明,使用MuMMI的性能和功率的预测错误率在大多数情况下小于10%。通过利用运行时、节点功率、CPU功率和内存功率的模型,我们确定了潜在应用程序优化的最重要性能计数器,并预测了优化的理论结果。基于两个收集的数据集,我们分析和比较使用MuMMI和10种机器学习方法在性能和功耗方面的预测精度。
Energy‐efficient scientific applications require insight into how high performance computing system features impact the applications' power and performance. This insight can result from the development of performance and power models. In this article, we use the modeling and prediction tool MuMMI (Multiple Metrics Modeling Infrastructure) and 10 machine learning methods to model and predict performance and power consumption and compare their prediction error rates. We use an algorithm‐based fault‐tolerant linear algebra code and a multilevel checkpointing fault‐tolerant heat distribution code to conduct our modeling and prediction study on the Cray XC40 Theta and IBM BG/Q Mira at Argonne National Laboratory and the Intel Haswell cluster Shepard at Sandia National Laboratories. Our experimental results show that the prediction error rates in performance and power using MuMMI are less than 10% for most cases. By utilizing the models for runtime, node power, CPU power, and memory power, we identify the most significant performance counters for potential application optimizations, and we predict theoretical outcomes of the optimizations. Based on two collected datasets, we analyze and compare the prediction accuracy in performance and power consumption using MuMMI and 10 machine learning methods.
IBM Blue Gene/Q 系统上的应用程序级功耗和性能表征及优化
DOI: 10.1147/jrd.2012.2227580
发表时间: 2013
期刊: IBM J. Res. Dev.
影响因子: --
作者:
Ramon Bertran Monfort;Yutaka Sugawara;H. Jacobson;A. Buyuktosunoglu;P. Bose
通讯作者: P. Bose
多级检查点的能量和运行时间之间的权衡分析
DOI: 10.1007/978-3-319-17248-4_13
发表时间: 2014
期刊: J. Parallel Distributed Comput.
影响因子: --
作者:
Prasanna Balaprakash;L. Bautista;M. Bouguerra;Stefan M. Wild;F. Cappello;P. Hovland
通讯作者: P. Hovland
DOI: 10.1109/dsnw.2012.6264670
发表时间: 2012-06
期刊: IEEE/IFIP International Conference on Dependable Systems and Networks Workshops (DSN 2012)
影响因子: --
作者:
M. Diouri;Olivier Glück;L. Lefèvre;F. Cappello
通讯作者: M. Diouri;Olivier Glück;L. Lefèvre;F. Cappello
DOI: 10.1145/2063384.2063427
发表时间: 2011-11
期刊: 2011 International Conference for High Performance Computing, Networking, Storage and Analysis (SC)
影响因子: --
作者:
L. Bautista-Gomez;S. Tsuboi;D. Komatitsch;F. Cappello;N. Maruyama;S. Matsuoka
通讯作者: L. Bautista-Gomez;S. Tsuboi;D. Komatitsch;F. Cappello;N. Maruyama;S. Matsuoka
PoLiMEr:适用于 HPC 应用的能源监控和功率限制接口
DOI: 10.1145/3149412.3149419
发表时间: 2017
期刊: Proceedings of the 5th International Workshop on Energy Efficient Supercomputing
影响因子: --
作者:
I. Marincic;V. Vishwanath;H. Hoffmann
通讯作者: H. Hoffmann