A data model for supplying a Data Center with several energy sources - PowerPoint PPT Presentation

A data model for supplying a Data Center with several energy sources Inès de Courchelle Tom Guérout Yann Labit Thierry Monteil IRIT/LAAS LAAS-CNRS LAAS-CNRS LAAS-CNRS idecourc@irit.fr tguerout@laas.fr ylabit@laas.fr monteil@laas.fr 18 juillet 2016

Plan I. Introduction II. RenewSim III. Modeling IV. Results V. Conclusion 2 26

Conclusion Introduction Introduction RenewSim Modeling Results CONTEXT ◆ Smart cities ◆ Home automation ◆ Use renewable energy sources ◆ Smart building ◆ Reducing the CO2 foot print ◆ Smart Grid ◆ ... ◆ Smart Building ◆ Supply a Data Center ◆ The running cost is more and more expensive ◆ Actual Data Center capacities ◆ 40 000 servers ◆ 500 000 services (VM) ◆ > 1 million 3 26

Conclusion Introduction Introduction RenewSim Modeling Results Data Center Consumption 2008 2020 2000 56 billion kW 104 billion kW in 70 Twh in the in European area European area world How to supply Data Center ? Economically ? 2011 2007 Environmentally ? France is the 330 Twh, 2% World C02 6 th electricity consumer production in the world 4 26

Conclusion Introduction Introduction RenewSim Modeling Results Challenge & Problematic ◆ Use several energy sources ◆ Control ◆ Green energy (solar panels, batteries, etc.) ◆ Energy Sources ◆ Brown energy (suppliers, etc.) ◆ Energy Destinations The problematic is how to characterize the heterogeneous electricity sources to make a decision to reduce electricity bills ? 5 26

Conclusion Introduction Introduction RenewSim Modeling Results Overview ◆ How is it composed ? ◆ What is it ? ◆ Each element is plugged to the bus ◆ Green IT Simulator ◆ Converter module ◆ Co-simulation tool ◆ Control system ◆ Based on a micro-grid architecture ◆ Why ? Context ◆ Use several energy sources ◆ Managing the power flow in a clever way 6 26

Conclusion Introduction Introduction RenewSim Modeling Results Control module ◆ Find the best trade-off (electricity vs IT) ◆ Manage the power flow into the smart grid ◆ Task allocation characterize by ◆ CPU & Memory capacities ◆ Energy demand and available servers RenewSim aims at reducing the energy cost by studying the power flow and the Data Center demand in a decision block. 7 26

Conclusion Introduction Introduction RenewSim Modeling Results The Modeling ◆ Manage the different degrees of freedom ◆ Specify the intrinsic behavior of each block 8 26

Conclusion Introduction Introduction RenewSim Modeling Results Freedom degrees 1/2 ◆ Freedom degrees ◆ The Suppliers ◆ The Batteries Context ◆ No degrees ◆ The power asked by the Data Center does not let us a lot freedom degrees ◆ Input of the system asked by the user and application ◆ Solar production does not depend on the system ◆ weather and time of the day 9 26

Conclusion Introduction Introduction RenewSim Modeling Results Freedom degrees 2/2 ◆ Batteries ◆ Store energy from solar panels or suppliers ◆ Unload energy from batteries ◆ The suppliers ◆ Price depends of time (night and day) ◆ Store energy into the batteries or sell it 10 26

Conclusion Introduction Introduction RenewSim Modeling Results Element characterization Source Gives energy to destinations (S t ) Destination Receives energy from the sources (D t ) Overflow Stores energy which can not be accepted by the destinations (O t ) Backup Unloads energy which can be feed the destinations (B t ) 11 26

Conclusion Introduction Introduction RenewSim Modeling Results Example (1/3) ◆ The modeling is presented as a matrix with 4 columns ◆ Each line is a component of our system Source Destination Backup Overflow Solar Panels 1 0 0 0 Data Center 0 1 0 0 Battery 0 0 1 1 Supplier 0 0 2 2 12 26

Conclusion Introduction Introduction RenewSim Modeling Results Example (2/3) 1 5 0 Source 5 0 2 5 5 0 0 0 0 0 0 Backup 2 Backup 1 Destination Overfull 2 Overfull 1 13 26

Conclusion Introduction Introduction RenewSim Modeling Results Example (3/3) 1 5 0 Source 5 0 2 5 0 0 0 0 0 0 Backup 2 Backup 1 Destination Overfull 2 Overfull 1 14 26

Conclusion Introduction Introduction RenewSim Modeling Results Algorithms 15 26

Conclusion Introduction Introduction RenewSim Modeling Results Study area (1/2) ◆ ADREAM ◆ Data solar panels: 48h ◆ 1648 measures ◆ Converter ◆ CM1: 15% ◆ CM2: 17% ◆ CM3: 13% ◆ CM4: 14% ◆ Inputs 16 26

Conclusion Introduction Introduction RenewSim Modeling Results Inputs (1/2) 17 26

Conclusion Introduction Introduction RenewSim Modeling Results Inputs (2/2) 18 26

Conclusion Introduction Introduction RenewSim Modeling Results Study area (2/2) Battery Load unload Capacity 10 000 joules Supplier Sale Purchase d 19 26

Conclusion Introduction Introduction RenewSim Modeling Results Experimentation 1 Source Destination Backup Overflow Solar Panels 1 0 0 0 Data Center 0 1 0 0 Battery 0 0 1 1 Supplier 0 0 2 2 20 26

Conclusion Introduction Introduction RenewSim Modeling Results Battery and supplier 21 26

Conclusion Introduction Introduction RenewSim Modeling Results Experimentation 2 Day Source Destination Backup Overflow Solar Panels 1 0 0 0 Data Center 0 1 0 0 Battery 1 0 1 1 Supplier 0 0 2 2 Night Source Destination Backup Overflow Solar Panels 1 0 0 0 Data Center 0 1 0 0 Battery 0 1 0 0 Supplier 1 0 0 0 22 26

Conclusion Introduction Introduction RenewSim Modeling Results Battery and supplier 23 26

Conclusion Introduction Introduction RenewSim Modeling Results Conclusion ◆ Renewsim allows: ◆ Design a Green modular architecture ◆ Simulate various kind of inputs and storage models ◆ Manage the power redirection through labeled and prioritized energy sources ◆ Analyze the impact of the IT feed on energy sources 24 26

Conclusion Introduction Introduction RenewSim Modeling Results Perspectives ◆ The expansion prospects consist in: ◆ Heuristics, algorithms of learning machine ◆ Prediction of production & usage ◆ Tasks priority ◆ QoS management ◆ The purpose of RenewSim is to be a co-simulation tool allowing communication simulators dedicated to each component. 25 26

Any Questions ? Inès de Courchelle Tom Guerout IRIT/LAAS LAAS-CNRS idecourc@irit.fr tguerout@laas.fr Yann Labit Thierry Monteil LAAS-CNRS LAAS-CNRS ylabit@laas.fr monteil@laas.fr

27 26

28 26

Conclusion Results Introduction Introduction Introduction Introduction Related Work RenewSim Micro-grid Modeling Modeling Results Example 5 0 Context 5 0 1 1 5 5 0 0 0 0 0 0 0 29 26