Data Center Analysis and Design Analysis of SU s Green Data Center Performance

Transcription

1 Data Center Analysis and Design Analysis of SU s Green Data Center Performance Dustin W. Demetriou, PhD Development Engineer Advanced Thermal Energy Efficiency Lab IBM D. W. Demetriou October 29,

2 Outline Why CoGen for Data Centers? New Metrics for CoGen Data Centers SU s Green Data Center Performance D. W. Demetriou October 29,

3 Data Center Power Supply Chain Opportunity for efficiency improvement with on-site cogeneration US DOE Industrial Technologies Program D. W. Demetriou October 29,

4 Why Cogeneration for Data Centers? With on-site power generation, the waste heat from the power system can be used to cool the data center, and cool/heat adjacent buildings. Because the electric and thermal demands of data centers are strongly correlated, the utilization of the cogeneration system s energy outputs can be much higher. On-site power generation can provide DC power to the data center, where almost all computer loads are DC. This reduces wasteful multiple conversions of AC/DC. Redundancy and uninterruptible power supply are potentially less expensive to provide in a data center powered and cooled by a cogeneration system. D. W. Demetriou October 29,

5 Typical Cogeneration System Block Diagram D. W. Demetriou October 29,

6 The SU/IBM Green Data Center Powered and cooled by an on-site CCHP system: 12 Capstone C65 micro-turbines 2 waste-heat activated double-effect absorption chillers Dual mode AC/DC power supply Grid connected or grid independent operation Battery backup (no conventional UPS) Propane fuel emergency backup Extensive use of water cooled racks Surplus cooling/heating/power is used by adjacent office building Goal: 50% reduction in primary energy use. D. W. Demetriou October 29,

7 Electric and Thermal Energy Flows D. W. Demetriou October 29,

8 Capstone C65 Microturbines Two banks of 6 turbines each provide N+1 reliability and generate a net 600 kw of electrical power. D. W. Demetriou October 29,

9 Turbine Exhaust Heat The exhaust of the turbines is ducted to the heat exchangers and the absorption chillers. D. W. Demetriou October 29,

10 Thermax Absorption Chillers 2 double-effect absorption chillers convert the 585 F exhaust into chilled water (~300 tons of cooling). D. W. Demetriou October 29,

11 Waste Heat Exchangers These heat exchangers provide domestic hot water and space heating to an adjacent campus building. D. W. Demetriou October 29,

12 Battery Backup 40 tons of sealed wet cell batteries provide ~20 minutes of runtime at full power. D. W. Demetriou October 29,

13 Non-Condensing Water Heat Exchangers These heat exchangers provide chilled water above the data center dew point temperature for the Rear Door HX and Sidecar HX. D. W. Demetriou October 29,

14 Water Cooling at the Rack Each of the equipment racks is water cooled. Rear-door HX removes 10kW+; Sidecar HX removes 30 kw+. D. W. Demetriou October 29,

15 IBM DS8300, Z10, and P575 The IBM Z10 Mainframe shown here (2 nd and 3 rd racks) is running on DC power. D. W. Demetriou October 29,

16 Typical Data Center Energy Flow Block Diagram ERE E IT E E IS IT Q EX Not thermodynamically or economically equivalent! E IT Electric Grid E DC = E IT +E IS Adjacent Building Q EX PUE EHeat IT EIS Recovery E IT D/C IT Load D/C Other Load Furnace With an on-site power system, a considerable amount of excess thermal energy can be recovered from the waste heat Chiller Gas or Oil Environment D. W. Demetriou October 29,

17 How to Fix TGG s Metric Part of the waste heat from the data center, Q HB, will displace heat that would have been supplied by a gas-fired furnace with an efficiency η f ; typically η f Part of the waste heat from the data center, Q CB, will displace cooling that would have been supplied by an electricity-driven chiller The ERE definition becomes: ERE 1 1 E IS Q CB /COP C E IT c g c e Q HB /η f where c g and c e are the gas and electricity prices in $/MJ. Therefore, C r (c g /c e )/η f The commercial gas/electricity price ratio in the US is typically , therefore a value of C r 0.35 is a reasonable average value. D. W. Demetriou October 29,

18 Multiple Operating Modes at SU GDC Data Center P G Grid P D Q CD T [P T ] Q W AbC [Q C ] Heater [Q H ] P TB Q CB Q HB Grid Gas Chiller [COP C ] Boiler [ f] Adjacent Building D. W. Demetriou October 29,

19 Performance Analysis (Fall 2011) 8pm until 7am D. W. Demetriou October 29,

20 Performance Analysis (Spring 2012) 1am until 11am D. W. Demetriou October 29,

21 Performance Analysis (Summer 2012) 12 am until 12 pm D. W. Demetriou October 29,

22 D. W. Demetriou October 29,

23 Backup Charts Multi-City Assessment of CoGen for Data Centers Metrics for CoGen Data Centers SU Performance Data D. W. Demetriou October 29,