v v Utilizing Green Energy Prediction to Adapt to Energy Supply Variability when Scheduling Mixed Batch and Service Jobs in Data Centers Renewable energy prediction • Instantaneous renewable energy usage may result in power shortages • Use short term prediction algorithms (30min window) to decide when renewable energy to run additional processing requests in order to achieve higher energy efficiency while meeting performance constraints Motivation • Data centers consume a lot of power • Millions of MWh, reflected as billions of dollars in the electricity bills • Tons of carbon emissions to the atmosphere • Idea: Use renewable energy to reduce carbon emissions • Problem: Renewable energy supply is highly variable which decreases the energy usage efficiency. Data center power consumption [Hitachi Eco-friendly Data center project, http://www.hitachi.com] Baris Aksanli, Jagannathan Venkatesh, Liuyi Zhang and Tajana Rosing UC San Diego NSF Expedition in Computing, Variability-Aware Software for Efficient Computing with Nanoscale Devices http://variability.org • EWMA -> 32.6 % error • Extended EWMA -> 23.4% error • Our work: WCMA -> 9.6% error • Combination of a weighted nearest-neighbor (NN) tables and wind power curve models • 21.2% error vs. state-of-the- art predictor 48.2% error System Architecture • Use green energy to schedule more batch jobs – MapReduce jobs are good as 92% finish within 30min • Terminate MR task when green energy supply drops • Guarantee service jobs meet their performance constraints, and limit performance hit to batch job completion times Results • Prediction has 15% lower job completion time on average compared to instantaneous case • Prediction has 2x overall higher GE efficiency vs. instantaneous • GE Efficiency: % of the GE that is used for useful work • Prediction leads to ~2x more jobs completed with green energy vs. instantaneous case • GE Job Percentage: ratio of jobs completed with GE over all completed jobs • On average, 5x fewer batch tasks need to be terminated with prediction. • %Incomplete Jobs: ratio of terminated jobs over all jobs completed with GE Comparison of solar prediction algorithms. Data gathered from solar installation at UCSD • Comparison of wind prediction algorithms. Data gathered from a wind farm in Lake Benton, available by NREL Baris Aksanli, Jagannathan Venkatesh, Liuyi Zhang and Tajana Rosing. Utilizing Green Energy Prediction to Schedule Mixed Batch and Service Jobs in Data Centers. International Workshop on Power Aware Computing and Systems, HotPower, 2011.

Utilizing Green Energy Prediction to Adapt to Energy ...seelab.ucsd.edu/virtualefficiency/resources/HotPower_11_poster.pdfUtilizing Green Energy Prediction to Adapt to Energy Supply

Download PDF Report

Upload
others
View
4
Download
0

Embed Size (px)

Citation preview

Page 1: Utilizing Green Energy Prediction to Adapt to Energy ...seelab.ucsd.edu/virtualefficiency/resources/HotPower_11_poster.pdfUtilizing Green Energy Prediction to Adapt to Energy Supply

v v

Utilizing Green Energy Prediction to Adapt to Energy Supply Variability when Scheduling Mixed

Batch and Service Jobs in Data Centers

Renewable energy prediction

• Instantaneous renewable energy usage may result in power shortages • Use short term prediction algorithms (30min window) to decide when renewable energy to run additional

processing requests in order to achieve higher energy efficiency while meeting performance constraints

Motivation • Data centers consume a lot of power

• Millions of MWh, reflected as billions of dollars in the electricity bills • Tons of carbon emissions to the atmosphere

• Idea: Use renewable energy to reduce carbon emissions • Problem: Renewable energy supply is highly variable which decreases

the energy usage efficiency. Data center power consumption [Hitachi Eco-friendly

Data center project, http://www.hitachi.com]

Baris Aksanli, Jagannathan Venkatesh, Liuyi Zhang and Tajana Rosing UC San Diego

NSF Expedi t ion in Comput ing, Var iab i l i ty -Aware Sof tware for Eff ic ient Comput ing wi th Nanoscale Devices h t tp : / /var iab i l i ty.org

• EWMA -> 32.6 % error • Extended EWMA -> 23.4% error •Our work: WCMA -> 9.6% error

•Combination of a weighted nearest-neighbor (NN) tables and wind power curve models •21.2% error vs. state-of-the-

art predictor 48.2% error

System Architecture

• Use green energy to schedule more batch jobs – MapReduce jobs are good as 92% finish within 30min

• Terminate MR task when green energy supply drops • Guarantee service jobs meet their performance

constraints, and limit performance hit to batch job completion times Results

•Prediction has 15% lower job completion time on average compared to instantaneous case

•Prediction has 2x overall higher GE efficiency vs. instantaneous • GE Efficiency: % of the GE

that is used for useful work

•Prediction leads to ~2x more jobs completed with green energy vs. instantaneous case • GE Job Percentage: ratio of

jobs completed with GE over

all completed jobs

•On average, 5x fewer batch tasks need to be terminated with prediction. • %Incomplete Jobs: ratio

of terminated jobs over all

jobs completed with GE

Comparison of solar prediction algorithms. Data gathered from solar installation at UCSD • Comparison of wind prediction algorithms.

Data gathered from a wind farm in Lake Benton, available by NREL

Baris Aksanli, Jagannathan Venkatesh, Liuyi Zhang and Tajana Rosing. Utilizing Green Energy Prediction to Schedule Mixed Batch and Service Jobs in Data Centers. International Workshop on Power Aware Computing and Systems, HotPower, 2011.

http://variability.org/