Predicting Energy Consumption by Extending the Palladio Component - - PowerPoint PPT Presentation

▶
predicting energy consumption by extending the palladio
SMART_READER_LITE
LIVE PREVIEW

Predicting Energy Consumption by Extending the Palladio Component - - PowerPoint PPT Presentation

Jun 02, 2023 456 likes •700 views

Stuttgart, Germany, 2014-11-27 Predicting Energy Consumption by Extending the Palladio Component Model Symposium on Software Performance (SOSP) 2014 Felix Willnecker 1 , Andreas Brunnert 1 , Helmut Krcmar 2 1 fortiss GmbH, 2 Technische

slide-1
SLIDE 1

fortiss GmbH An-Institut Technische Universität München Stuttgart, Germany, 2014-11-27

Predicting Energy Consumption by Extending the Palladio Component Model

Symposium on Software Performance (SOSP) 2014 Felix Willnecker1, Andreas Brunnert1, Helmut Krcmar2

1fortiss GmbH, 2Technische Universität München

slide-2
SLIDE 2

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 2

  • Motivation
  • Power Consumption Model

– Calculation – Meta-Model Extension – Power Consumption Model Generation

  • Evaluation

– SPECjEnterprise 2010 – Runtastic for Android

  • Conclusion

Agenda

slide-3
SLIDE 3

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 3

  • Motivation
  • Power Consumption Model

– Calculation – Meta-Model Extension – Power Consumption Model Generation

  • Evaluation

– SPECjEnterprise 2010 – Runtastic for Android

  • Conclusion

Agenda

slide-4
SLIDE 4

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 4

  • Energy Consumption of Information and

Communication Technology is growing (Stobbe et al. 2009, Willnecker et al. 2014)

  • Optimization on hardware and operating system

level cannot compensate rising demand (Gottschalk et al. 2012)

  • Investigating the energy efficiency on application

level becomes a growing software engineering challenge (Brunnert et al. 2014)

  • Main challenges and goals:

– Reduce operation costs in data centers – Increase battery life time of portable devices – Ease carbon footprint

Motivation

slide-5
SLIDE 5

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 5

  • Performance metrics and energy consumption rely on the same underlying

parameters of resource demand and hardware capabilities

  • Performance simulation and prediction techniques can be used to predict

the energy consumption of applications

Motivation

Performance Model Simulation Hardware Workload Software Response time Throughput Resource demand

slide-6
SLIDE 6

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 6

  • Performance metrics and energy consumption rely on the same underlying

parameters of resource demand and hardware capabilities

  • Performance simulation and prediction techniques can be used to predict

the energy consumption of applications

Motivation

Performance Model Simulation Energy consumption Hardware Workload Software

slide-7
SLIDE 7

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 7

  • Motivation
  • Power Consumption Model

– Calculation – Meta-Model Extension – Power Consumption Model Generation

  • Evaluation

– SPECjEnterprise 2010 – Runtastic for Android

  • Conclusion

Agenda

slide-8
SLIDE 8

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 8

Calculation

  • This function is calculated for each resource container or linking resource

based on the utilization of the components in this resource container.

  • Ppred is the Predicted Power Consumption of a resource container
  • Pidle,0 is the Power Consumption function of the resource container in idle

state

  • Pidle,i is the Power Consumption factor function of a component in this

resource container

  • ui is the utilization factor of the component (e.g, CPU utilization, GPS on/off

state, throughput)

Power Consumption Model

slide-9
SLIDE 9

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 9

Meta-Model Extension

Power Consumption Model

slide-10
SLIDE 10

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 10

Meta-Model Extension

Power Consumption Model

Resource Container

slide-11
SLIDE 11

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 11

Meta-Model Extension

Power Consumption Model

Linking Resource

slide-12
SLIDE 12

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 12

Power Consumption Model of a server

Power Consumption Model

  • Ppred = 200 + 300 x uCPU + 50 x uHDD
  • The energy consumption E of the system is the

integral over Ppred over the simulation time T

Server in PCM with Power Consumption Model (Brunnert et al. 2014b)

slide-13
SLIDE 13

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 13

Power Consumption Model of a mobile device

Power Consumption Model

  • Battery capacity specified

to calculate discharging

  • Stochastic functions for

varying power consumptions

  • Added two resource types:

GPS and DISPLAY

  • Linking resource power

consumption model to calculate energy demand of network traffic based on throughput

Mobile Device in PCM with Power Consumption Model

slide-14
SLIDE 14

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 14

Power Consumption Model Generation

  • Manually creation based on specifications and

estimations

– Resource Specifications from manufacturer – Android Vendor Profiles

  • Calibration by stressing resources independently

– Intelligent Plattform Management Interface – Android Calibration App

Power Consumption Model

slide-15
SLIDE 15

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 15

  • Motivation
  • Power Consumption Model

– Calculation – Meta-Model Extension – Power Consumption Model Generation

  • Evaluation

– SPECjEnterprise 2010 – Runtastic for Android

  • Conclusion

Agenda

slide-16
SLIDE 16

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 16

SPECjEnterprise 2010

Evaluation

System Under Test (Brunnert et al. 2014b) IBM System X3755M3 Virtual Server (VM Ware ESXi 5.0.0)

  • penSuse 12.3

Benchmark Driver Load Balancer Load Driver AMD-based Server Intel-based Server IBM System X3550M3

  • penSuse 12.3

DB JBoss Application Server JBoss Application Server IBM System X3755M3

  • penSuse 12.2

DB JBoss Application Server JBoss Application Server

slide-17
SLIDE 17

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 17

SPECjEnterprise 2010

Evaluation

Clients MMPC1 SMPC2 PCPE3 1300 367.55 W 320.26 W 12.87 % 2300 403.87 W 352.22 W 12.79 % 3300 433.76 W 384.52 W 11.35 % 3500 436.47 W 390.95 W 10.43 % Clients MMPC1 SMPC2 PCPE3 1300 197.05 W 175.94 W 10.71 % 2300 220.47 W 194.93 W 11.58 % 3300 241.67 W 213.91 W 11.49 % 4300 264.29 W 232.69 W 11.96 %

1 Measured Mean Power Consumption 2 Simulated Mean Power Consumption 3 Power Consumption Prediction Error

Measured and simulated results for the AMD-based server (Brunnert et al. 2014b) Measured and simulated results for the Intel-based server (Brunnert et al. 2014b)

slide-18
SLIDE 18

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 18

Runtastic for Android

Evaluation

  • Nexus 5

running Android 4.4

  • Galaxy Tab

running Android 4.3

  • Runtastic

running on both devices 30 mins run

Nexus 5 Galaxy Tab MMPC1 0.883 W 1.251 W SMPC2 0.732 W 1.084 W PCPE3 17.12 % 13.35 % BLPE4 0.67 % 1.01 %

1 Measured Mean Power Consumption 2 Simulated Mean Power Consumption 3 Power Consumption Prediction Error 4 Battery Level Prediction Error

Measured and simulated results for the mobile devices (Leimhofer 2014)

slide-19
SLIDE 19

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 19

  • Motivation
  • Power Consumption Model

– Calculation – Meta-Model Extension – Power Consumption Model Generation

  • Evaluation

– SPECjEnterprise 2010 – Runtastic for Android

  • Conclusion

Agenda

slide-20
SLIDE 20

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 20

  • Energy consumption can be predicted using the Palladio Component Model

with an error of mostly below 13% for server systems and 17,2 % for mobile devices.

  • Multi-Processor Environments for mobile hard to calibrate
  • Extension for other device types (Windows, iOS, etc.)
  • Additional resource types (e.g., accelerometer) are necessary for mobile

device evaluations

  • Power Consumption Model Repository for different devices from different

vendors.

  • Automatic Performance Model Generation for mobile devices.

Conclusion

slide-21
SLIDE 21

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 21

Brunnert, A.; Vögele, C.; Danciu, A.; Pfaff, M.; Mayer, M.; Krcmar, H. (2014): Performance Management Work. In: Business & Information Systems Engineering, Vol. 6 (2014), pp. 1-3. Brunnert, A.; Wischer, K.; Krcmar, H. (2014): Using Architecture-Level Performance Models as Resource Profiles for Enterprise Applications. In: Proceedings of the 10th ACM SIGSOFT International Conference on the Quality of Software Architectures (QoSA), Lille, France. Gottschalk, M.; Josefiok, M.; Jelschen, J.; Winter, A. (2012): Removing Energy Code Smells with Reengineering

  • Services. Paper presented at the GI-Jahrestagung, Braunschweig, Germany, pp. 441-455.

Leimhofer, J. (2014): Predicting the Energy Consumption of Mobile Applications using Simulations. Bachelor's Thesis, Technische Universität München 2014. Stobbe, L.; Nissen, N.; Proske, M.; Middendorf, A.; Schlomann, B.; Friedewald, M.; Georgieff, P.; Leimbach, T. (2009): Abschätzung des Energiebedarfs der weiteren Entwicklung der Informationsgesellschaft. In: Abschlussbericht an das Bundesministerium für Wirtschaft und Technologie. Berlin, Karlsruhe: Fraunhofer IZM, (2009). Willnecker, F.; Brunnert, A.; Krcmar, H. (2014): Model-based Energy Consumption Prediction for Mobile

  • Application. In: Proceedings of the 28th Conference on Environmental Informatics - Informatics for Environmental

Protection, Sustainable Development and Risk Management, (2014).

Bibliography

slide-22
SLIDE 22

pmw.fortiss.org SOSP 2014, Stuttgart, Germany, 2014-11-27 22

Q&A

Felix Willnecker, Andreas Brunnert

performancegroup@fortiss.org pmw.fortiss.org