PerfEstimator: A Generic and Extensible Performance Estimator for Data Parallel DNN Training
Understanding the performance of data parallel DNN training at large-scale is crucial for supporting efficient DNN cloud deployment as well as facilitating the design and optimization of scalable DNN systems. Existing works adopt analytical modeling, which may fall short in capturing the system behaviors resulting from the fast evolving DNN systems and constantly proposed optimizations. In this paper, we present PerfEstimator, a generic and extensible estimator for accurate performance estimation of large-scale data parallel DNN training. PerfEstimator is driven by three major components, namely, an extensible attributed graph based performance model, a computation and synchronization profiling and simulating tool for obtaining runtime time costs on a single machine, and a computation-synchronization pipeline builder to derive the scaling factors. Our evaluation highlights that PerfEstimator can accurately predict the performance of data parallel DNN training jobs with a prediction error of 0.2-11%.
Sat 29 MayDisplayed time zone: Amsterdam, Berlin, Bern, Rome, Stockholm, Vienna change
11:55 - 12:55 | Technical paper session #1CloudIntelligence 2021 at CloudIntelligence Room Chair(s): Qingwei Lin Microsoft Research, Beijing, China | ||
11:55 15mPaper | PerfEstimator: A Generic and Extensible Performance Estimator for Data Parallel DNN Training CloudIntelligence 2021 Chengru Yang University of Science and Technology of China, Zhehao Li University of Science and Technology of China, Chaoyi Ruan University of Science and Technology of China, Guanbin Xu University of Science and Technology of China, Cheng Li University of Science and Technology of China, Ruichuan Chen Nokia Bell Labs, Feng Yan University of Nevada Reno | ||
12:10 15mPaper | Kmon: An In-kernel Transparent Monitoring System for Microservice Systems with eBPF CloudIntelligence 2021 Tianjun Weng Sun Yat-Sen University, Wanqi Yang Sun Yat-Sen University, Guangba Yu Sun Yat-Sen University, Pengfei Chen Sun Yat-Sen University, Jieqi Cui Sun Yat-Sen University, Chuanfu Zhang Sun Yat-Sen University | ||
12:25 15mPaper | TraceLingo: Trace representation and learning for performance issue diagnosis in cloud services CloudIntelligence 2021 Yong Xu Microsoft, China, Yaokang Zhu Microsoft Research Asia, Bo Qiao Microsoft Research, Beijing, China, Hongshu Che Microsoft Research, Beijing, China, Pu Zhao Microsoft Research, Beijing, China, Xu Zhang Microsoft Research, Beijing, China, Ze Li Microsoft, USA, Yingnong Dang Microsoft, USA, Qingwei Lin Microsoft Research, Beijing, China | ||
12:40 15mPaper | MicroDiag: Fine-grained Performance Diagnosis for Microservice Systems CloudIntelligence 2021 Li Wu Elastisys AB/Technische Universität Berlin, Johan Tordsson Elastisys AB, Jasmin Bogatinovski , Erik Elmroth Elastisys AB/Umea University, Odej Kao Technische Universität Berlin | ||
Go directly to this room on Clowdr