Increasing the Efficiency of UPS Systems – And Proving It! Richard L. Sawyer Director, Critical Facilities Assurance EYP Mission Critical Facilities www.eypmcf.com

The Problem 60% of US Energy bill is in buildings. Energy consumed by data centers more than doubled between 2000 and 2005 – J. Koomey, Stanford University. U.S. Data center electrical bills totaled $2.7 Billion in 2005. A single, moderate size server in a data center has the same carbon foot print as a SUV that gets 15 MPG (R.Muirhead, Data Center Journal). A single rack with 6 Blade Server units consumes as much power as 3 kitchen electric ranges (24-30Kw)!

Relative Power Densities

21st Century Computing – Blade Servers Power = Up to 6 kW per Blade chassis or 30 kW per rack

Where does the power go? UPS = 18% Actual IT Load is 30% of Power Consumed APC-MGE: Neil Rasmussen

UPS OUTPUT POWER INPUT POWER FROM UTILITY/GENERATOR DISTORTION SPIKE SWELL SAG OUTAGE Lightning Strikes Faulty Switchgear Storms High Winds Falling Trees Traffic Accidents Heavy Loads Poor Distribution Switching Operations Poor Filters Faulty Load Eq. Static Electricity RF Interference Harmonics/ Electronic Loads FREQUENCY Major Utility Problems Faulty Generator INPUT POWER FROM UTILITY/GENERATOR UPS OUTPUT POWER PURPOSES OF UNINTERRUPTIBLE POWER SUPPLY 1.Maintain clean, uninterrupted power during utility events 2.Power Conditioning 3.Isolation from other electrical loads 4.Separately Derived Source of Power

Strategy to Improve UPS Efficiency IBM Blue Gene 1.2 Megawatt Technology: Make the units more efficient. Selection: Size the units more closely to the load. Application: Use redundancy only where it is needed.

Understanding UPS Inefficiency Factors No-Load Losses Proportional Losses Square-Law Losses Paying the price to process power!

EPRI Efficiency Curves for UPS Products

Typical UPS efficiency curve Below 30% load efficiency drops rapidly Nominal 92% efficiency only applies when UPS load is over 70% 100% 90% 80% 70% 60% UPS Efficiency 50% 40% 30% 20% 10% 0% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% UPS Load % of full power rating

13.5 KV 13.5 KV 2(N+1) System 480 480 Each side must have capacity to support both critical loads but maintain redundancy. Total load cannot exceed capacity of 2 UPS Modules. EFFECTIVE DESIGN LOAD = 33% of total capacity, maximum. Primary Bus A Primary Bus B UPS UPS UPS UPS UPS UPS Bypass A Bypass B Load Bank Load Bank UPS Output 2A UPS Output 2B Subsystem Bus A Subsystem Bus B Critical Load Bus A Critical Load Bus B Static Switch Static Switch PDU PDU Critical Load

Aggregate UPS Power Losses No-load portion of loss stays constant from full load all the way down to zero load { } No-load loss is present even at no load Many data centers operate in this range UPS load % of full power rating 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Power delivered to load UPS internal power consumption (loss) 93.4% 93.3% 93.1% 92.8% 92.4% 91.8% 90.7% 88.9% 85.5% 76.4% EFFICIENCY Proportional and square losses

No Load Losses Definition: The power consumed by the UPS at 0% load just to keep the UPS operating. Sources – Transformers, capacitors, logic systems, fans, communications cards. Sometimes referred to as “tare”, “constant”, “fixed”, “shunt” and “parallel” losses. Most significant inefficiency: Accounts for up to 40% of UPS losses.

Proportional Losses Definition: The power needed to process more power through the UPS. Sources – Switching losses, capacitor and inductor impedance, internal resistance Proportional losses increase as the output load the UPS support increases. Proportional losses are directly related to the topology (internal design) of the UPS.

Square - Law Losses Definition: Losses related to the amount of current flowing through the UPS. Power is the result of voltage times the current. Current does the work, and power is lost as the amount of current flowing increases, by a square factor, hence “square – law losses”. Power loss is in the form of heat. Square-Law losses are 1% to 4% at higher load levels.

Power Loss Component Graph No Load Electrical Loss in kW (Waste due to inefficiency) Equipment Loading Full Load 50% 10% 30% 90% 70% NO-LOAD loss PROPORTIONAL loss SQUARE-LAW loss TOTAL LOSS

(Waste due to inefficiency) Two devices with same nameplate efficiency can have significantly different losses in actual operating range, due to the particular characteristics of their PROPORTIONAL and NO-LOAD losses Example: Two different 100kW UPSs with 92% nameplate (full-load) efficiency Same nameplate efficiency (full-load loss) 10kW Loading where most data centers operate Electrical Loss (Waste due to inefficiency) UPS A TOTAL LOSS UPS B has higher proportional loss (steeper line) but lower no-load loss UPS B TOTAL LOSS UPS A No-load loss UPS B No-load loss But different performance at actual operating load 0kW No Load 10% 30% 50% 70% 90% Full Load Equipment Loading

10kW 0kW Electrical Loss Equipment Loading One device can even have WORSE nameplate efficiency than another, yet have lower loss in actual operating range, if it has a low NO-LOAD loss Example: Two 100kW UPSs with same 92% nameplate (full-load) efficiency UPS A has better nameplate efficiency (lower full-load loss) 10kW Loading where most data centers operate B A Electrical Loss (Waste due to inefficiency) UPS A TOTAL LOSS UPS B has higher proportional loss (steeper line) but lower no-load loss UPS B TOTAL LOSS UPS A No-load loss UPS B No-load loss But UPS B performs better at actual operating load 0kW No Load 10% 30% 50% 70% 90% Full Load Equipment Loading

Technology Selection Application Improving Efficiency Technology Selection Application

Improving Efficiency – Fixing No-Load Loss Effect of lowering NO-LOAD LOSS Example: 100kW UPS with 92% full-load efficiency 10kW Nameplate efficiency goes from 92% to 94.5% Same improvement in nameplate efficiency Loading where most data centers operate Total loss before improvement Electrical Loss (Waste due to inefficiency) Total loss after improvement Electric bill savings Original No-load loss But waste is roughly cut in half in actual operating range Lowered No-load loss 0kW No Load 10% 30% 50% 70% 90% Full Load Equipment Loading

Improving Efficiency – Fixing Proportional Loss Effect of lowering PROPORTIONAL LOSS Example: 100kW UPS with 92% full-load efficiency 10kW Nameplate efficiency goes from 92% to 94.5% Loading where most data centers operate Total loss before improvement Electrical Loss (Waste due to inefficiency) Total loss after improvement Electric bill savings (Unchanged No-load loss) Waste is reduced by 10-20% in actual operating range 0kW No Load 10% 30% 50% 70% 90% Full Load Equipment Loading

Application Efficiency – Zoned Redundancy M CRAC pdu UPS Cold Aisle Hot F I R E S C U HEAT REJECT EPO SYSTEM MONITOR WEBLINK Site Availability – 99.995% $2,000+ per square foot Battery Central UPS for one “N” side, scalable, modular system Rack Based UPS Systems as needed for 2N redundancy

Commissioning UPS Systems Availability The Cost of Downtime The Value of Commssioning

Data Center Tier Ratings * The Uptime Institute

Maximizing Availability Total Time - Downtime Availability = Total Time The only variable is Downtime Downtime sources: Equipment Failures, Human Error, External Causes, Maintenance Cost of Downtime drives the Value of CFA!

What does Downtime Cost?

Infant Mortality Period High Probability of Downtime The Reliability Curve for equipment (IEEE) Infant Mortality Period End-of- Life Period High Probability of Downtime Failure Rate Time (Data Center Life Span) “The Bathtub Curve”

Infant Mortality Period The Value of Commissioning Infant Mortality Period End-of- Life Period Minimize Failures Time

Commissioning UPS Systems Verify the full load performance of each module using load banks – typical burn in is 4 hours at rated KW load (hint: infrared inspections of all connections). Measure and verify the efficiency in the full operating range at 5%, 10%, 15%, 20%, 25%.......... Verify system redundancy under design load levels. Verify failure modes (under-voltage transfers, bypass transfers, over load shutdown). Verify isolation modes for concurrent maintenance. Assuring you get the reliability and efficiency you pay for!

Richard L. Sawyer 518-337-2049 rsawyer@eypmcf.com Questions? Richard L. Sawyer 518-337-2049 rsawyer@eypmcf.com