DC Power for Data Centers – demonstration summary My Ton – Ecos Consulting Brian Fortenbery – EPRI Solutions Bill Tschudi – Lawrence Berkeley National Laboratory Sponsored by: California Energy Commission (CEC)─Public Interest Energy Research (PIER), California Institute for Energy Efficiency (CIEE).
Overview – Phase 1 Rationale for the study Background on power conversions and their efficiencies Demonstration objectives Industry partners Configurations Results
Thomas Edison: “My personal desire would be to prohibit entirely the use of alternating currents. They are as unnecessary as they are dangerous. I can therefore see no justification for the introduction of a system which has no element of permanency and every element of danger to life and property.”
California Energy Commission Public Interest Energy Research High-tech Buildings Project Objectives • Research, develop, and demonstrate, innovative energy efficient technologies • 10-year initiative focusing on high-tech industries – e.g. data centers • Help move the market to more efficient technologies • Research and demonstration projects include technology transfer
Why look at Data Centers? • Data center power use nationally is large and growing. • Two prior studies estimated data center energy use: • 2004 EPRI/Ecos estimated 14.8 TWh • 2000 Arthur D. Little estimated 10.1 TWh • 0ne terawatthour = 1,000,000,000 kilowatthours or • one million megawatthours • Saving a fraction of this energy is substantial
Loads Power delivery Cooling Cumulative Power Representative Data Center Power Use ~50% Power Efficiency 6 Source: Intel Corp.
Power to meet a 100 W Computing Load Load 100W Total 275W VR 20W PSU 50W Server fans 15W UPS +PDU 20W Room cooling system 70W source: Intel Corporation Source: Intel Corp. 7
DC Demonstration – Timeline • Stakeholders first met – Fall 2005 • Kick-off meeting – April 2006 • Equipment assembly – May 2006 • Initial “Team Open House” June 7, 2006 • Public Open House events: June 21, July 12, 26; Aug 9, 16 • End date – August 16, 2006
Industry Partners Made it Happen • Alindeska Electrical Contractors • APC • Baldwin Technologies • Cisco Systems • Cupertino Electric • Dranetz-BMI • Emerson Network Power • Industrial Network Manufacturing (IEM) Equipment and Services Contributors: • Intel • Nextek Power Systems • Pentadyne • Rosendin Electric • SatCon Power Systems • Square D/Schneider Electric • Sun Microsystems • UNIVERSAL Electric Corp.
Other Partners Collaborated • 380voltsdc.com • CCG Facility Integration • Cingular Wireless • Dupont Fabros • EDG2, Inc. • EYP Mission Critical • Gannett • Hewlett Packard Stakeholders: • Morrison Hershfield Corporation • NTT Facilities • RTKL • SBC Global • TDI Power • Verizon Wireless
Data Center Power Use • Data center power use nationally is large and growing. • Two studies estimated data center energy use: • 2004 EPRI/Ecos estimated 14.8 TWh • 2000 Arthur D. Little estimated 10.1 TWh • 0ne terawatthour = 1,000,000,000 kilowatthours or • one million megawatthours • Saving a fraction of this energy is substantial
This demonstration focused on reducing power delivery and conversion losses observed in our prior work: Power Supplies in IT equipment Uninterruptible Power Supplies (UPS)
UPS and Power Supply efficiency • We observed a wide range of performance from the worst to the best • Our original goal was to move the market to the higher performing systems • Incentive programs, labeling, education programs were all options – and still are
Data Center Power Delivery System DC/DC 78 - 85% UPS 88 - 92% Power Dist 98 - 99% Power Supply 68 - 72% The heat generated from the losses at each step of power conversion requires additional cooling power HVAC: Power for cooling can equal or exceed the direct losses
The questions we were addressing: Could some of the conversion steps be eliminated to improve efficiency? Could a demonstration be devised to measure actual savings?
DC Demonstration - Objectives The demonstration’s original objectives were to show a rack level solution: • DC powered server equipment exists in the same form factor or can readily be built from existing components • DC powered server equipment can provide the same level of functionality and computing performance when compared to similarly configured and operating AC server equipment • Efficiency gains from the elimination of multiple conversion steps can be measured by comparing traditional AC delivery to a DC system • DC system reliability could be as good or better than AC system reliability
The project team soon defined additional objectives: • Demonstration of 380 V. DC distribution at the facility level compared to conventional AC systems • Demonstration of other DC solutions (48 volt systems) • Evaluation of safety considerations • Demonstrate ability to connect alternative energy solutions (PV, fuel cells, etc.)
What the demonstration included • Side-by-side comparison of traditional AC system with new DC system • Facility level distribution • Rack level distribution • Power measurements at conversion points • Servers modified to accept 380 V. DC • Artificial loads to more fully simulate data center
Additional items included • Racks distributing 48 volts to illustrate that other DC solutions are available, however no energy monitoring was provided for this configuration • DC lighting was included!
Facility-Level DC Distribution 380V.DC
Details • Safety was reviewed by a committee of the partners. No significant issues were identified. Only concern was whether fault currents would be large enough to trip protective devices. Final report will address safety and applicable codes and standards • All distribution equipment is UL rated for DC applications • No standard connector has been agreed upon for the server DC connection • With widespread adoption, reliability should be improved – fewer potential points of failure. Eliminating heat sources should help.
Measured Results • Facility level overall efficiency improvement: 10 to 20% • Smaller rack level overall efficiency improvement but other benefits include: • Thermal benefits • Smaller power supply in server • Transition strategy for existing centers
AC system loss compared to DC 9% measured improvement 2-5% measured improvement
Implications could be even better for a typical data center • Redundant UPS and server power supplies operate at reduced efficiency • Cooling loads would be reduced. • Both UPS systems used in the AC base case were “best in class” sytems and performed better than benchmarked systems – efficiency gains compared to typical systems could be higher. • Further optimization of conversion devices/voltages is possible
UPS XFMR PS Total Efficiency System Efficiency 87.00% 98.00% 90.00% 76.73% High Efficiency (DC Option) 92.00% 100.00% 92.00% 84.64% Compute Load (W) Input Load (W) Difference System Load 10,000 13032.03 High Efficiency (DC Option) 10,000 11814.74 9.34% Data Center Power Delivery System UPS 87 - 92% XFMR 98% - NA Power Supply 90 - 92%
UPS XFMR PS Total Efficiency Typical System Efficiency 85.00% 98.00% 73.00% 60.81% High Efficiency (DC Option) 92.00% 100.00% 82.00% 75.44% Optimized DC Option 92.00% 100.00% 92.00% 84.64% Compute Load (W) Input Load (W) Difference Typical Load 10,000 16444.93 High Efficiency (DC Option) 10,000 13255.57 19.39% Optimized DC Option 10,000 11814.74 28.16% Data Center Power Delivery System UPS 85 - 92% Power Dist 98% - NA Power Supply 73 - 92%
Results What does 15% increase in efficiency mean to the electrical power grid?
DC Power - next steps: • DC power pilot installation(s) • Standardize distribution voltage • Standardize DC connector and power strip design • Server manufacturers develop power supply specifications • Power supply manufacturers develop prototypes • UL and communications certification • Address other types of IT equipment (storage, switches, etc.)
Follow progress on DC power in data centers on-line Lawrence Berkeley National Laboratory websites for more information • http://hightech.lbl.gov/ • http://hightech.lbl.gov/dc-powering/
Additional Information Project Coordination & Contacts: Lawrence Berkeley National Laboratory • Bill Tschudi, Principal Investigator email@example.com Ecos Consulting • My Ton firstname.lastname@example.org EPRI Solutions • Brian Fortenbery email@example.com