Energy Efficient Data Centers. Update on LBNL data center energy efficiency projects. June 23, 2005 Bill Tschudi Lawrence Berkeley National Laboratory Wftschudi@lbl.gov. LBNL’s energy research related to data centers. Energy research roadmap Case studies and energy benchmarking
Energy Efficient Data Centers Update on LBNL data center energy efficiency projects June 23, 2005 Bill Tschudi Lawrence Berkeley National Laboratory Wftschudi@lbl.gov
LBNL’s energy research related to data centers • Energy research roadmap • Case studies and energy benchmarking • Best practice identification • Self benchmarking protocol • Investigate efficiency of power supplies in IT equipment • Investigate efficiency of UPS systems • Metrics for computing performance vs. energy • Technology transfer • Demonstration projects
Data center efficiency opportunity • Many efficiency ideas have been identified through industry feedback • Case studies are helping to identify best practices
Data center efficiency resources • ASHRAE “Thermal Guidelines for Data Processing Environments” • ASHRAE “Power Trends and Cooling Applications” • In preparation: ASHRAE “Design Considerations for Data Center and Communications Equipment Centers” which includes a chapter on energy efficiency
Case studies/benchmarks • California • Storage device and router Mfgs. • Banks • Web hosting facilities • Internet service provider • State tax center • Federal facilities • New York • Recovery center (hosting) • Financial institution
IT equipment load intensities Data collected in 1999 through 2003 showed that electrical power intensity for IT equipment alone was on the order of 25 Watts/sf. Current data suggests that load intensities are rising through compaction and/or due to rising equipment power consumption.
2003 IT equipment loads from LBNL case studies Average 27 +/- (W/sf of electrically active floor space using Uptime definition)
2003 projections if fully loaded Projected Average 44 (W/sf of electrically active floor space using Uptime definition)
2005 IT equipment benchmarks LBNL NERSC supercomputer Average 52 w/sf
Electrical power conversion is a big opportunity Every conversion of AC voltage, AC to DC, DC to AC, or DC voltage conversion results in loss of electrical power and corresponding heat that must be removed from the data center. Minimizing this conversion loss has a magnifying effect that allows all facility systems to use less energy and consequently the infrastructure systems can be downsized. Saving say 10% on the conversion loss could result in 20% or more saving for the facility.
5V Internal Drive PWM/PFC Switcher Unregulated DC To Multi Output Regulated DC Voltages Bypass 12V External Drive 3.3V I/O 12V 1.5/2.5V In Out DC/DC Memory Controller AC/DC 1.1V-1.85V 12V DC/DC m Processor Battery/Charger Rectifier Inverter 3.3V SDRAM 3.3V Graphics Controller Voltage Regulator Modules AC/DC Multi output PS How many times do data centers convert AC and DC? AC voltage conversions
Electricity use in a server Based on a typical dual processor 450W 2U Server; Approximately160W out of 450W (35%)are losses in the power conversion process (Source: Brian Griffith: INTEL)
Power supply opportunity Based on one case study approximately 4335 KW of a total of 8500 kW was IT load. Assuming a 65% existing baseline efficiency, the savings opportunity using 90% efficient conversion process is approximately 1300kWnot including any savings from HVAC
Power supply efficiency recommendations The Server Systems Infrastructure group (SSI) publishes recommended minimum efficiencies for server power supplies. The LBNL project team is working with this group to see what can be done to raise the bar.
Energy efficiency opportunity Specifiers of UPS’s or IT equipment can have a huge impact on energy use by requiring higher efficiencies. Testing data shows that higher efficiencies can be obtained – you have to ask for it. Facility and IT professionals by working together can optimize overall power conversions. Additional costs (if any) for more efficient conversions will have a very short payback or may be entirely justified by reductions in infrastructure.
Index of performance The Uptime Institute proposed a metric to evaluate the total efficiency of infrastructure systems: Index of performance = building systems KW UPS output (i.e. ratio of building systems to IT equipment load)
Look at the end-use Data Center A Data Center B
LBNL high-tech buildings website: http://hightech.lbl.gov