Please wait a minute...
浙江大学学报(工学版)
J4  2012, Vol. 46 Issue (4): 725-733    DOI: 10.3785/j.issn.1008-973X.2012.04.022
    
DVFS-aware CPU service time estimation method
ZHANG Zhen, LI Shan-ping
College of Computer Science and Technology, Zhejiang University, Hangzhou 310029, China
Download:   PDF(0KB) HTML
Export: BibTeX | EndNote (RIS)      

Abstract  

A service time estimation method that use the product of average frequency and CPU utilization instead of CPU utilization as the dependent variables of regression analysis was proposed in order to mitigate the large error due to the ignorance of CPU dynamic voltage and frequency scaling (DVFS) during existing CPU service time estimation methods. The cpufreq_stats driver of Linux was modified to accurately measure the average CPU frequency, and the problem that the original driver underestimates average frequency was fixed. A method to revise existing frequency readings was proposed for the environment that patching cpufreq_stats is not possible. Experiments with a micro benchmark application in Linux show that DVFS can significantly impact the estimated service time of the classic regression method. For the services with small service time, it can cause around 100% deviation, while the DVFS-aware regression method can still give accurate estimation. Various average frequency measurement approaches were compared. Results show that current tools can not give accurate average frequency, and the relative error can be larger than 40%.

Published: 17 May 2012
CLC:  TP 302.7  
Service
E-mail this article
Add to my bookshelf
Add to citation manager
E-mail Alert
RSS
Articles by authors
Cite this article:

ZHANG Zhen, LI Shan-ping. DVFS-aware CPU service time estimation method. J4, 2012, 46(4): 725-733.

URL:

http://www.zjujournals.com/eng/10.3785/j.issn.1008-973X.2012.04.022     OR     http://www.zjujournals.com/eng/Y2012/V46/I4/725


变频感知的处理器服务时间估算方法

为了减少现有处理器服务时间估算方法忽略处理器动态电压和频率调节造成的较大误差,提出将平均频率和处理器利用率的乘积替代处理器利用率作为回归分析中因变量的服务时间估算方法.为了准确地测量平均频率,修改了Linux操作系统的cpufreq_stats驱动,修复了原有驱动低估平均频率的问题.针对无法安装该补丁的环境,提出修正现有频率读数的方法.通过在Linux下微基准测试程序的实验,发现动态频率调节能够显著影响采用经典方法估算的服务时间；对于部分服务时间较小的服务,会造成100%左右的误差,而采用变频敏感的回归方法能够准确地估计处理器服务时间.比较各种平均频率测量方法,结果显示现有测量工具无法给出准确的平均频率,相对误差可以达到40%以上.

［1］ ZHANG Q, CHERKASOVA L, MATHEWS G, et al.RCapriccio: a capacity planning and anomaly detection tool for enterprise services with live workloads［C］∥Proceedings of the ACMIFIPUSENIX 2007 International Conference on Middleware. Newport Beach: SpringerVerlag, 2007: 244-265.
［2］ ZHANG Q, CHERKASOVA L, MI N, et al.A regressionbased analytic model for capacity planning of multitier applications［J］. Cluster Computing,2008, 11(3): 197-211.
［3］ URGAONKAR B, PACIFICI G, SHENOY P, et al. Analytic modeling of multitier Internet applications ［J］. ACM Transactions on the Web, 2007,1(1): article 2.
［4］ BARHAM P, DONNELLY A, ISAACS R, et al.Using magpie for request extraction and workload modelling［C］∥Proceedings of the 6th Conference on Symposium on Opearting Systems Design and Implementation. San Francisco: USENIX, 2004: 18.
［5］ CHEN M Y, KICIMAN E, FRATKIN E, et al. Pinpoint: problem determination in large, dynamic Internet services ［C］∥Proceedings of the 2002 International Conference on Dependable Systems and Networks. Los Alamitos: IEEE, 2002: 595-604.
［6］ ROLIA J, VETLAND V.Correlating resource demand information with ARM data for application services ［C］∥Proceedings of the 1st International Workshop on Software and Performance. Santa Fe: ACM,1998: 219-230.
［7］ ZHANG L, XIA C H, SQUILLANTE M S, et al.Workload service requirements analysis: a queueing network optimization approach［C］∥Proceedings of 10th IEEE International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunications Systems. Washington DC: IEEE, 2002: 23-32.
［8］ STEWART C, KELLY T, ZHANG A.Exploiting nonstationarity for performance prediction［C］∥Proceedings of the 2nd ACM SIGOPS/EuroSys European Conference on Computer Systems. Lisbon: ACM,2007: 31-44.
［9］ CHERKASOVA L, OZONAT K, MI N, et al.Automated anomaly detection and performance modeling of enterprise applications ［J］. ACM Transactions on Computer Systems, 2009, 27(3): 1-32.
［10］ PACIFICI G, SEGMULLER W, SPREITZER M, et al.CPU demand for web serving: measurement analysis and dynamic estimation［J］. Performance Evaluation, 2008, 65(6/7): 531-553.
［11］ MENASCE D A, DOWDY L W, ALMEIDA V A F. Performance by design: computer capacity planning by example［M］. Upper Saddle River: Prentice Hall, 2004: 64-65.
［12］ PALLIPADI V, STARIKOVSKIY A.The ondemand governor: past, present and future ［C］∥Proceedings of Linux Symposium. Ottawa: \
[s. n.\], 2006: 223-238.
［13］ DRAPER N R, SMITH H. Applied regression analysis ［M］.2nd ed. New York: Wiley, 1981: 417.
No related articles found!