Multi-objective negotiation of power profiles for datacenter powered with renewable energies Léo Grange University of Toulouse Institut de Recherche en Informatique de Toulouse (IRIT) July 2018
Context and overview Approach Methodology and evaluation Conclusion Contextandoverview (IRIT- University of Toulouse) GreenDays@Toulouse 2018 1 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Datacenter consumption and renewable sources Worldwide: 270 TWh in 2012 ≈ Italy electricity consumption High economical and environmental costs Possible mitigations Improve energy efficiency, software and hardware Use renewable energy sources power Solar, wind: intermittent and little predictability New challenges to make efficient use in datacenters ANR Datazero: on-site renewable sources IT and electrical cooperation (IRIT- University of Toulouse) GreenDays@Toulouse 2018 2 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Electrical infrastructure IT infrastructure Forecast Power profile Power profile evaluation evaluation PDM ITDM Optimization algorithm Tasks Separated IT and electrical optimizations Power Ability to evaluate power plan impact 12 10 Internal objective (utility) 8 Black box functions R T → R 6 4 Computationally expensive 2 0 Time 0:00 6:00 12:00 18:00 (IRIT- University of Toulouse) GreenDays@Toulouse 2018 3 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Electrical infrastructure IT infrastructure Forecast Power profile Power profile evaluation evaluation PDM ITDM Optimization algorithm Tasks Separated IT and electrical optimizations Power Ability to evaluate power plan impact 12 10 Internal objective (utility) 8 Black box functions R T → R 6 4 Computationally expensive 2 0 Time 0:00 6:00 12:00 18:00 (IRIT- University of Toulouse) GreenDays@Toulouse 2018 3 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Electrical infrastructure IT infrastructure Forecast Power profile Power profile evaluation evaluation PDM ITDM Optimization algorithm Tasks Separated IT and electrical optimizations Electrical utility Power Ability to evaluate power plan impact 12 10 Internal objective (utility) 8 Black box functions R T → R 6 4 Computationally expensive 2 0 Time 0:00 6:00 12:00 18:00 (IRIT- University of Toulouse) GreenDays@Toulouse 2018 3 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Electrical infrastructure IT infrastructure Forecast Power profile Power profile evaluation evaluation PDM ITDM Optimization algorithm Tasks Separated IT and electrical optimizations Electrical utility IT utility Power Ability to evaluate power plan impact 12 10 Internal objective (utility) 8 Black box functions R T → R 6 4 Computationally expensive 2 0 Time 0:00 6:00 12:00 18:00 (IRIT- University of Toulouse) GreenDays@Toulouse 2018 3 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Multi-objective aspect Each DM has one or more objectives to satisfy Objectives may differ between DM QoS related for ITDM, environmental impact for PDM Managing different objectives Avoiding the problem: find common objective (money?) Scalarization (e.g. weighted sum) Finding a set of good solutions (set of possible trade-offs) (IRIT- University of Toulouse) GreenDays@Toulouse 2018 4 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Multi-objective aspect Each DM has one or more objectives to satisfy Objectives may differ between DM QoS related for ITDM, environmental impact for PDM Managing different objectives Avoiding the problem: find common objective (money?) Scalarization (e.g. weighted sum) Finding a set of good solutions (set of possible trade-offs) (IRIT- University of Toulouse) GreenDays@Toulouse 2018 4 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Multi-objective aspect Each DM has one or more objectives to satisfy Objectives may differ between DM QoS related for ITDM, environmental impact for PDM Managing different objectives Avoiding the problem: find common objective (money?) Scalarization (e.g. weighted sum) Finding a set of good solutions (set of possible trade-offs) (IRIT- University of Toulouse) GreenDays@Toulouse 2018 4 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Multi-objective optimization and heuristics Electrical utility 15 10 5 0 Find Pareto front 5 (best trade-offs) 10 15 20 50 0 50 100 150 IT utility Multi-Objective Evolutionary Algorithms Well studied area, various approaches Focused on SPEA2 (genetic algorithm) (IRIT- University of Toulouse) GreenDays@Toulouse 2018 5 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Multi-objective optimization and heuristics Electrical utility 15 10 5 0 Find Pareto front Power 5 (best trade-offs) 10 0:00 12:00 0:00 15 Time 20 50 0 50 100 150 IT utility Multi-Objective Evolutionary Algorithms Well studied area, various approaches Focused on SPEA2 (genetic algorithm) (IRIT- University of Toulouse) GreenDays@Toulouse 2018 5 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Multi-objective optimization and heuristics Electrical utility 15 Power 10 5 0:00 12:00 0:00 Time 0 Find Pareto front Power 5 (best trade-offs) 10 0:00 12:00 0:00 15 Time 20 50 0 50 100 150 IT utility Multi-Objective Evolutionary Algorithms Well studied area, various approaches Focused on SPEA2 (genetic algorithm) (IRIT- University of Toulouse) GreenDays@Toulouse 2018 5 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Context and overview of the problem Multi-objective optimization and heuristics Electrical utility 15 Power 10 5 0:00 12:00 0:00 Time 0 Find Pareto front Power 5 (best trade-offs) 10 0:00 12:00 0:00 15 Time 20 50 0 50 100 150 IT utility Multi-Objective Evolutionary Algorithms Well studied area, various approaches Focused on SPEA2 (genetic algorithm) (IRIT- University of Toulouse) GreenDays@Toulouse 2018 5 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Utility approximation Approximation of power profile utility Evaluation of power profile is costly Genetic algorithms require many evaluations Workaround: Utility approximation Fast approximation based on known solutions Evaluate only potentially good ones Example: utility function, 2 time steps PDM utility for 2 time slots 6 0.8 0.4 5 0.0 1st time slot power 4 0.4 3 0.8 2 1.2 1 1.6 0 2.0 0 1 2 3 4 5 6 2nd time slot power (IRIT- University of Toulouse) GreenDays@Toulouse 2018 6 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Utility approximation Approximation of power profile utility Evaluation of power profile is costly Genetic algorithms require many evaluations Workaround: Utility approximation Fast approximation based on known solutions Evaluate only potentially good ones Example: utility function, 2 time steps PDM utility for 2 time slots 6 0.8 6 0.4 5 0.0 1st time slot power 4 4 Power 0.4 3 2 0.8 2 1.2 0 0 1 2 1 1.6 Time step u = − 0 . 29 0 2.0 0 1 2 3 4 5 6 2nd time slot power (IRIT- University of Toulouse) GreenDays@Toulouse 2018 6 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Utility approximation Approximation of power profile utility Evaluation of power profile is costly Genetic algorithms require many evaluations Workaround: Utility approximation Fast approximation based on known solutions Evaluate only potentially good ones Example: utility function, 2 time steps PDM utility for 2 time slots 6 0.8 6 0.4 5 0.0 1st time slot power 4 4 Power 0.4 3 2 0.8 2 1.2 0 0 1 2 1 1.6 Time step u = − 0 . 29 0 2.0 0 1 2 3 4 5 6 2nd time slot power (IRIT- University of Toulouse) GreenDays@Toulouse 2018 6 / 21 Léo Grange
Context and overview Approach Methodology and evaluation Conclusion Utility approximation Approximation of power profile utility Evaluation of power profile is costly Genetic algorithms require many evaluations Workaround: Utility approximation Fast approximation based on known solutions Evaluate only potentially good ones Example: utility function, 2 time steps PDM utility for 2 time slots 6 0.8 6 0.4 5 0.0 1st time slot power 4 4 Power 0.4 3 2 0.8 2 1.2 0 0 1 2 1 1.6 Time step u = 0 . 41 0 2.0 0 1 2 3 4 5 6 2nd time slot power (IRIT- University of Toulouse) GreenDays@Toulouse 2018 6 / 21 Léo Grange
Recommend
More recommend
