User- and Process-Driven Dynamic Voltage and Frequency Scaling Bin Lin Arindam Mallik Peter Dinda Gokhan Memik Robert Dick Empathic Systems Project Department.

Slides:



Advertisements
Similar presentations
Chapter 14: Usability testing and field studies
Advertisements

Arindam Mallik Jack Cosgrove Robert P. Dick Gokhan Memik Peter Dinda Northwestern University Department of Electrical Engineering and Computer Science.
R2: An application-level kernel for record and replay Z. Guo, X. Wang, J. Tang, X. Liu, Z. Xu, M. Wu, M. F. Kaashoek, Z. Zhang, (MSR Asia, Tsinghua, MIT),
In this presentation you will:
Improving TCP Performance over Mobile Ad Hoc Networks by Exploiting Cross- Layer Information Awareness Xin Yu Department Of Computer Science New York University,
Techniques for Multicore Thermal Management Field Cady, Bin Fu and Kai Ren.
A Cyber-Physical Systems Approach to Energy Management in Data Centers Presented by Chen He Adopted form the paper authors.
Chapter 14: Usability testing and field studies. 2 FJK User-Centered Design and Development Instructor: Franz J. Kurfess Computer Science Dept.
The User In Experimental Computer Systems Research Peter A. Dinda Gokhan Memik, Robert Dick Bin Lin, Arindam Mallik, Ashish Gupta, Sam Rossoff Department.
NuCAD ELECTRICAL ENGINEERING AND COMPUTER SCIENCE McCormick Northwestern University Robert R. McCormick School of Engineering and Applied Science FA-STAC.
A Stratified Approach for Supporting High Throughput Event Processing Applications July 2009 Geetika T. LakshmananYuri G. RabinovichOpher Etzion IBM T.
Performance and Energy Bounds for Multimedia Applications on Dual-processor Power-aware SoC Platforms Weng-Fai WONG 黄荣辉 Dept. of Computer Science National.
1 Drafting Behind Akamai (Travelocity-Based Detouring) AoJan Su, David R. Choffnes, Aleksandar Kuzmanovic, and Fabian E. Bustamante Department of Electrical.
1 Razor: A Low Power Processor Design Presented By: - Murali Dharan.
System-Wide Energy Minimization for Real-Time Tasks: Lower Bound and Approximation Xiliang Zhong and Cheng-Zhong Xu Dept. of Electrical & Computer Engg.
The User In Experimental Computer Systems Research Peter A. Dinda Gokhan Memik, Robert Dick Bin Lin, Arindam Mallik, Ashish Gupta, Sam Rossoff Department.
1 Dong Lu, Peter A. Dinda Prescience Laboratory Computer Science Department Northwestern University Virtualized.
Characterizing and Predicting TCP Throughput on the Wide Area Network Dong Lu, Yi Qiao, Peter Dinda, Fabian Bustamante Department of Computer Science Northwestern.
Power Reduction Through Measurement and Modeling of Users and CPUs Process-driven Voltage Scaling (PDVS) User-driven Frequency Scaling (UDFS) Bin Lin,
The User In Experimental Computer Systems Research Peter A. Dinda Gokhan Memik, Robert Dick Bin Lin, Arindam Mallik, Ashish Gupta, Sam Rossoff Department.
Administrator’s Guide
Into the Wild: Studying Real User Activity Patterns to Guide Power Optimizations for Mobile Architectures Alex Shye, Benjamin Scholbrock, and Gokhan Memik.
Power-Aware SoC Test Optimization through Dynamic Voltage and Frequency Scaling Vijay Sheshadri, Vishwani D. Agrawal, Prathima Agrawal Dept. of Electrical.
Alex Shye, Berkin Ozisikyilmaz, Arindam Mallik, Gokhan Memik, Peter A. Dinda, Robert P. Dick, and Alok N. Choudhary Northwestern University, EECS International.
VOLTAGE SCHEDULING HEURISTIC for REAL-TIME TASK GRAPHS D. Roychowdhury, I. Koren, C. M. Krishna University of Massachusetts, Amherst Y.-H. Lee Arizona.
XI HE Computing and Information Science Rochester Institute of Technology Rochester, NY USA Rochester Institute of Technology Service.
Dynamic Resource Allocation Using Virtual Machines for Cloud Computing Environment.
Computer Architecture and Operating Systems CS 3230: Operating System Section Lecture OS-3 CPU Scheduling Department of Computer Science and Software Engineering.
EmNet: Satisfying The Individual User Through Empathic Home Networks J. Scott Miller, John R. Lange & Peter A. Dinda Department of Electrical Engineering.
Presented By : SOMESH PAL (13IS23F) VISHAL BABU (13IS25F) MTECH CSE-IS NITK SURATHKAL Dynamic Reduction of Voltage Margins by Leveraging On-chip ECC in.
An Efficient Algorithm for Dual-Voltage Design Without Need for Level-Conversion SSST 2012 Mridula Allani Intel Corporation, Austin, TX (Formerly.
Games are Up for DVFS Yan Gu Samarjit Chakraborty Wei Tsang Ooi Department of Computer Science National University of Singapore.
System-level, Unified In-band and Out-of-band Dynamic Thermal Control Dong LiVirginia Tech Rong GeMarquette University Kirk CameronVirginia Tech.
Critical Power Slope Understanding the Runtime Effects of Frequency Scaling Akihiko Miyoshi, Charles Lefurgy, Eric Van Hensbergen Ram Rajamony Raj Rajkumar.
A Framework for Elastic Execution of Existing MPI Programs Aarthi Raveendran Graduate Student Department Of CSE 1.
Optimization Problems - Optimization: In the real world, there are many problems (e.g. Traveling Salesman Problem, Playing Chess ) that have numerous possible.
Shared Memory Parallelization of Decision Tree Construction Using a General Middleware Ruoming Jin Gagan Agrawal Department of Computer and Information.
1 Distributed Energy-Efficient Scheduling for Data-Intensive Applications with Deadline Constraints on Data Grids Cong Liu and Xiao Qin Auburn University.
The Fast Optimal Voltage Partitioning Algorithm For Peak Power Density Minimization Jia Wang, Shiyan Hu Department of Electrical and Computer Engineering.
Experiences with Client-based Speculative Remote Display John R. Lange & Peter A. Dinda Department of Electrical Engineering and Computer Science Northwestern.
Performance Prediction for Random Write Reductions: A Case Study in Modelling Shared Memory Programs Ruoming Jin Gagan Agrawal Department of Computer and.
Measuring Interactive Performance with VNCplay Nickolai Zeldovich, Ramesh Chandra Stanford University.
VGreen: A System for Energy Efficient Manager in Virtualized Environments G. Dhiman, G Marchetti, T Rosing ISLPED 2009.
Dynamic Voltage Frequency Scaling for Multi-tasking Systems Using Online Learning Gaurav DhimanTajana Simunic Rosing Department of Computer Science and.
An Energy-efficient Task Scheduler for Multi-core Platforms with per-core DVFS Based on Task Characteristics Ching-Chi Lin Institute of Information Science,
MROrder: Flexible Job Ordering Optimization for Online MapReduce Workloads School of Computer Engineering Nanyang Technological University 30 th Aug 2013.
VSched: Mixing Batch And Interactive Virtual Machines Using Periodic Real-time Scheduling Bin Lin Peter A. Dinda Prescience Lab Department of Electrical.
XI HE Computing and Information Science Rochester Institute of Technology Rochester, NY USA Rochester Institute of Technology Service.
Multimedia Computing and Networking Jan Reduced Energy Decoding of MPEG Streams Malena Mesarina, HP Labs/UCLA CS Dept Yoshio Turner, HP Labs.
Xi He Golisano College of Computing and Information Sciences Rochester Institute of Technology Rochester, NY THERMAL-AWARE RESOURCE.
Power Capping Via Forced Idleness ANSHUL GANDHI Carnegie Mellon Univ. 1.
1 Improved Policies for Drowsy Caches in Embedded Processors Junpei Zushi Gang Zeng Hiroyuki Tomiyama Hiroaki Takada (Nagoya University) Koji Inoue (Kyushu.
Chapter 4 CPU Scheduling. 2 Basic Concepts Scheduling Criteria Scheduling Algorithms Multiple-Processor Scheduling Real-Time Scheduling Algorithm Evaluation.
Jacob R. Lorch Microsoft Research
Empathic Computer Architectures and Systems
Frequency Governors for Cloud Database OLTP Workloads
Virtualization, Empathic Systems, and Sensors
International Symposium on Microarchitecture. New York, NY.
Alex Shye, Yan Pan, Ben Scholbrock, J. Scott Miller,
Empathic Computer Architectures and Systems
Miss rate versus (period, slice)
CPU Scheduling G.Anuradha
Learning and Leveraging the Relationship between Architecture-Level Measurements and Individual User Satisfaction Alex Shye, Berkin Ozisikyilmaz, Arindam.
Zhen Xiao, Qi Chen, and Haipeng Luo May 2013
Dynamic Voltage Scaling
Power improvement in the multitasking environment
Press ESC for Startup Options © Microsoft Corporation.
User-driven Scheduling Of Interactive Virtual Machines
COMP755 Advanced Operating Systems
Presentation transcript:

User- and Process-Driven Dynamic Voltage and Frequency Scaling Bin Lin Arindam Mallik Peter Dinda Gokhan Memik Robert Dick Empathic Systems Project Department of Electrical Engineering and Computer Science Northwestern University

2 Summary User-Driven Frequency Scaling (UDFS) –Adjust CPU frequency according to satisfaction of individual user, exploiting variation in satisfaction among users Process-Driven Voltage Scaling (PDVS) –Adjust voltage at frequency and temperature based on profile of the individual processor, exploiting variation among processors ~50% power reduction on Windows tasks compared to Windows DVFS –User study, measured system power

3 Outline Summary User-Driven frequency scaling (UDFS) –User pessimism study –UDFS2 algorithm Process-Driven voltage scaling (PDVS) –PDVS profiling Evaluation of UDFS and PDVS Work in context Conclusion

4 User Satisfaction With CPU Frequency How satisfied are users of different applications at different clock frequencies? Earlier work suggests there will much variation User Study –8 users –3 frequencies + Windows XP DVFS –3 typical Windows apps Presentation, Animation, Game –Rate comfort on Leikert-like 1 to 10 scale

5 Dramatic variation in user satisfaction for fixed frequencies –And for DVFS Setting frequency for worse case user is pessimistic Setting it for an “average” user is also wrong Presentation Game

6 User-Driven Frequency Scaling (UDFS) Developed system to dynamically customize frequency to the individual user –Key idea: user feedback at run-time –User presses “irritation button” (F11) as input But not too often! –2 very simple learning algorithms UDFS1, UDFS2

UDFS2 Algorithm Goal: probe to find the lowest frequency level the user is comfortable at and stabilize there General idea: react additively to both passage of time and user irritation events; also update rate of reaction based on rate of user feedback –UDSF1, in contrast, is an additive increase/multiplicative decrease algorithm, analogous to TCP Reno congestion control 7

UDFS2 Algorithm Each frequency level r i has interval t i –t i = 10 seconds, initially –r i are frequencies in descending order User not irritated in current interval t i ? –Switch frequency to r i+1, interval t i+1 Let’s go slower in the next interval 8

UDFS2 Algorithm User irritated in current interval t i ? –t i-1 =  t i-1 (  We should spend more time at the next highest frequency –t k =  t k  k: k  i-1 (  We should spend less time at the other frequencies, including this one –Switch frequency to r i-1, interval t i-1  = 2.5,  =0.8 in our studies 9

UDFS2 in Action (User is playing a game) 10 Note Convergence

11 Outline Summary User-Driven frequency scaling (UDFS) –User pessimism study –UDFS2 algorithm Process-Driven voltage scaling (PDVS) –PDVS profiling Evaluation of UDFS and PDVS Work in context Conclusion

CPU Voltage Minimum voltage needed for CPU stability is a function of frequency and temperature Claim: function varies across individual parts for many interesting processors Hence, we can do better than the worst case (nominal) function –P  V 2 CF makes even small changes significant 12

Example GHz Pentium M-770 in a Thinkpad T43p This processor can be run at lower voltages than specifications indicate

Process-Driven Voltage Scaling 14 Profiling System (Boot CD) –Workload generator –Temperature monitor –Frequency selector –Voltage selector –Watchdog timer For each frequency/temperature –Modulate workload to maintain temperature –Decrease voltage until watchdog timer fires and reboots machine Profiling system implemented by Nikolay Valtchanov and Matt Robben Profile: min_voltage(frequency, temperature) Any DVFS Scheme

15 Outline Summary User-Driven frequency scaling (UDFS) –User pessimism study –UDFS2 algorithm Process-Driven voltage scaling (PDVS) –PDVS profiling Evaluation of UDFS and PDVS Work in context Conclusion

Experimental Setup IBM Thinkpad T43p –2.13 GHz Pentium M-770 –1 GB RAM –Windows XP SP2 UDFS system (or Windows DVFS) runs online, adjusting CPU frequency as user interacts with the system and applications PDVS effects and power measurements/analysis are done offline, using logs from UDFS and user traces –System power measurement –CPU dynamic power via simulation 16

Application Tasks Use Microsoft Powerpoint 2003 to replicate a presentation while listening to background music with Windows Media Player 10 Watch 3D Shockwave animation with Microsoft Internet Explorer (locally stored animation) Play FIFA 2005 Soccer (first person shooter game) 17

Users 20 participants recruited from Northwestern population via IRB-approved methods Participants self-identified as “Power User”, “Typical User”, “Beginner” for each application, plus PCs and Windows A demographic mix 18

Study Process Fill out a questionnaire (2 minutes) Read a one page handout (2 minutes) Acclimatize to the performance of our machine by using the apps (5 minutes) Perform the following tasks for UDFS1 –Powerpoint (4 minutes) –3D Shockwave (4 minutes) –FIFA game (8 minutes) Repeat previous for UDFS2 19

Results Compared to Windows DVFS Unless Otherwise Noted Measured system power –UDFS1, UDFS2, UDFS1+PDVS, UDFS2+PDVS, Windows DVFS+PDVS Simulated CPU dynamic power –UDFS1, UDFS2, UDFS1+PDVS, UDFS2+PDVS, Windows DVFS+PDVS Measured mean and peak temperature –UDFS1, UDFS2, UDFS1+PDVS, UDFS2+PDVS User irritation event rate (UDFS1,2) Multitasking study 20

Measured System Power (PowerPoint, % improvement over Windows DVFS) 21 PDVS dominates gains for less interactive applications with little user-user variance

Measured System Power (FIFA, % improvement over Windows DVFS) 22 UDFS contributes significant gains for more interactive applications

Measured Temperature (PowerPoint) 23 Significant temperature reductions are possible using UDFS and PDVS

Measured Temperature (FIFA Game) 24 Significant temperature reductions are possible using UDFS and PDVS

User Irritation Button Presses Rate of user feedback can be low, and can decrease with time 25 First 4 minutesSecond 4 minutes

26 Outline Summary User-Driven frequency scaling (UDFS) –User pessimism study –UDFS2 algorithm Process-Driven voltage scaling (PDVS) –PDVS profiling Evaluation of UDFS and PDVS Work in context Conclusion

Work In Context of Power Management Efforts in the Empathic Systems Project (empathicsystems.org) A. Gupta, B. Lin, P. Dinda, Measuring And Understanding User Comfort With Resource Borrowing, HPDC 2004 –Opportunity paper: Identification of variance in user satisfaction with systems decisions P. Dinda, G. Memik, R. Dick, B. Lin, A. Mallik, A. Gupta, S. Rossoff, The User In Experimental Computer Systems Research, ExpCS 2007 –Position paper: overview of goals of project and advocacy for user-driven work A. Mallik, J. Cosgrove, R. Dick, G. Memik, P. Dinda, PICSEL: Measuring User-Perceived Performance to Control Dynamic Frequency Scaling, ASPLOS 2008 –Customize power management to user by observing output to him A. Shye, B. Ozisikyilmaz, A. Mallik, G. Memik, P. Dinda, R. Dick, A. Choudhary, Learning and Leveraging the Relationship between Architectural-level Measurements and Individual User Satisfaction, ISCA 2008 –Learn performance counter->user satisfaction and use in power management A. Shye, Y. Pan, B. Scholbrock, J. S. Miller, G. Memik, P. Dinda, R. Dick, Power to the People: Leveraging Human Physiological Traits to Control Microprocessor Frequency, MICRO 2008 –Measure user satisfaction using biometrics and use in power management [This paper] –Direct user feedback for power management + PDVS 27

Related Work DVFS in general –Gochman, et al [Intel Tech Journal], Broch, et al [SOC03], and many more… Dynamic Thermal Management –Liu, et al [IEEE JSSC-93], Brookes, et al [WCED00], Crusoe, Intel Pentium-M, etc… PDVS-related –Teodorescu, et al [ISCA08], Razor [Ernst, et al, MICRO03], Dhar, et al [ISLPED05], Intel Foxton, AutoDVS [EMSOFT05], etc… UDFS-related –Lorch, et al [UCB TR], Yan, et al [DAC05], Vertigo [OSDI02], Xu, et al [EMSOFT05], Ranga, et al [IEEE Computer-06], Anand, et al [MOBICOMM03], Theocharous, et al [Intel Tech Journal 06] 28

29 Conclusion User-Driven Frequency Scaling (UDFS) –Adjust CPU frequency according to satisfaction of individual user, exploiting variation in satisfaction among users Process-Driven Voltage Scaling (PDVS) –Adjust voltage at frequency and temperature based on profile of the individual processor, exploiting variation among processors ~50% power reduction on Windows tasks compared to Windows DVFS –User study, measured system power Variation is opportunity

30 For More Information Empathic Systems Project – Prescience Lab – Peter Dinda –

Measured System Power (Multitasking Study, % improvement over Windows DVFS)) 31 Multitasking appears to increase benefits Small additional study where user watched 3D animation while also listening to MP3s using Windows Media Player

Measured System Power (3D Shockwave, % improvement over Windows DVFS) 32 UDFS contributes significant gains for more interactive applications

Measured Temperature (3D Shockwave) 33 Significant temperature reductions are possible using UDFS and PDVS