1. Datacenter Energy Efficiency Research: An Update Lawrence Berkeley National Laboratory Bill Tschudi wftschudi@lbl.gov July 29, 2004

2. Data center energy efficiency research and deployment sponsors California Energy Commission Pacific Gas and Electric Co. New York State Energy Research and Development Authority DOE Federal Energy Management Program

3. Today’s outline Overview of past and present research Energy benchmarking and best practices Power supplies in IT equipment Efficient UPS systems Demonstrations Q & A

4. Completed activities OVERVIEW Data center case studies and benchmarking Research roadmap for CA CA market assessment

5. OVERVIEW Prior work in New York Case studies/energy benchmarking Workshops

6. OVERVIEW Related prior activities Jon Koomey study of impact of the Internet Jennifer Jackson-Mitchell thesis/case study Case studies/energy benchmarking Studies of standby losses and power supplies

7. OVERVIEW Current activities Benchmarking and best practices identification Self-benchmarking protocol Investigate uninterruptible power supplies (UPS) Investigate power supplies in IT equipment Collaborate on building interface developments

8. –NYSERDA sponsored project Air management – modeling and demonstration with SUNY, GA TECH, and IBM –CEETherm University of Maryland/Georgia Tech Related activities OVERVIEW

9. 6-10 data centers – (still seeking sites) Self-benchmarking protocol Review all benchmark results to determine better performing systems Best Practices Add benchmarks for standby generation Investigate other performance metrics Energy benchmarking Benchmarking

10. local distribution lines to the building, 480 V HVAC system lights, office space, etc. UPSPDUcomputer racks backup diesel generators Energy end use in data centers computer equipment uninterruptible load UPS = Uninterruptible Power Supply PDU = Power Distribution Unit;

11. “In God we trust – all others bring data” Quote: Lee Eng-Lock

12. Power consumption in data centers Benchmarking

13. Benchmarks of IT equipment load density Benchmarking

14. Projections if fully loaded Benchmarking

15. Distribution of computer room power reported to uptime institute Benchmarking

16. HVAC effectiveness Benchmarking

17. Benchmarks of chillers Benchmarking 7.5 average

18. Benchmarking UPS efficiency

19. Inverter InOut Bypass Battery/Charger Rectifier Internal Drive External Drive I/O Memory Controller  Processor SDRAM Graphics Controller DC/DC AC/DC DC/DC AC/DC Multi output PS Voltage Regulator Modules 5V 12V 3.3V 12V 1.5/2. 5V 1.1V- 1.85V 3.3V 12V PWM/PFC Switcher Unregulated DC To Multi Output Regulated DC Voltages How many times are we converting electricity? Typical existing data center configuration

20. Measure performance of various systems at several load factors Investigate effects of different redundancy configurations Develop energy efficiency spec for UPS systems Scoping study for data center DC architecture Investigating efficiency opportunity for UPS systems UPS Systems

21. Status of UPS investigation Protocols developed for field or shop measurement. Mfgs specifications reviewed for efficiency versus load for static and inertial UPS topology. Calculation tools developed: For comparing AC versus DC powering at rack level For evaluating cost savings due to higher efficiency UPS. Completed testing of UPS to show impact of “high efficiency” option on static UPS Coordinating with International labeling effort for power quality & efficiency UPS Systems

22. Measured efficiency vs. load Sample of 12 field measurements, including ten from LBNL roadmap activity UPS Systems

23. Potential savings UPS Systems Preliminary macro and rack level energy savings estimate for improving UPS efficiency.

24. Delta Conversion Double Conversion Delta conversion UPS efficiency manufacturers data Courtesy: American Power Conversion (APC)

25. Operational/installation/space considerations also have a role in designing the UPS configuration. If actual load is 50% of design load, each UPS is loaded 25% If actual load is 50% of design load, each UPS is loaded 33% Design Load = 600kW 600kW 300kW UPS configuration and impact on loading

27. Document efficiency of existing power supplies Measure performance to determine real life efficiency Recommend efficiency performance thresholds Work with Server System Infrastructure (SSI) group to advance better performance Investigating power supplies commonly used in IT equipment Energy Efficient Power Supplies

28. Electricity use in a server Based on a typical dual processor 450W 2U Server; Approximately 160W out of 450W (35%) are losses in the power conversion process (Source: Brian Griffith: INTEL) Energy Efficient Power Supplies

29. Recommending more efficient power supplies Existing Efficiency Recommendation in SSI PS Design Guides The Server System Infrastructure (SSI) Initiative’s goal is to deliver a set of specifications covering two primary server elements: power supplies and electronics bays. Members of SSI are Intel, Dell, HP, Silicon Graphics, IBM. Energy Efficient Power Supplies

30. Full Load Efficiency > 68% Redundant System of Power Supplies for Servers Energy Efficient Power Supplies

31. Testing power supplies

32. Percent Loading Factory Spec Test Results 100%94%89.6%* 85%N/A89.4% 50%86.5% Real life efficiency Based on EPRI PEAC Testing of a 20kVA, 3-Ph Double Conversion UPS Energy Efficient Power Supplies

33. Importance of a flat power supply efficiency curve Energy Efficient Power Supplies

34. Redundant Power Supply (300W + 300W) At 50% Design Load each supply is loaded 25% 1200 W 4+1 (N+1) Redundant Power Supply System At 50% design load each supply is loaded 40% Power supply system redundancy can be accomplished without using “redundant power supplies”. Power supply redundancy Energy Efficient Power Supplies

35. Courtesy: American Power Conversion (APC) 80% of the time the processor usage was less than 4% of its peak capability A GHz processor is using MHz resource most of the time

36. Looking for production metrics Very Low Processor Activity does not relate to very low power consumption Most of the time the GHz processor is doing activities that can be done by a MHz processor but the input power consumption does not change much

37. CA public utilities’ emerging technology program and the CA Energy Commission are planning demonstration projects. Possibilities: –Air management –Air flow visualization –Alternative cooling strategies –UPS upgrades and reconfigurations –Rack level DC distribution –On site power generation –Ideas? Demonstration projects

38. Technology Transfer LBNL data center websites http://hightech.lbl.gov/ http://Datacenters.lbl.gov

40. Thank you Questions?