Материал (02 юни 2010)

Emerson Power Network

Choosing the right UPS for small and midsize data centers: a cost and reliability comparison


A dependable uninterruptible power supply (UPS) system is essential to protectingdata centers and server rooms from unplanned downtime and equipment damage.

This paper compares two UPS systems marketed for small and midsize data centers.These products represent remarkably different design philosophies. The impact ofthese differing philosophies on reliability and cost of ownership is addressed.In addition, this paper addresses the following questions:

  • What is the appropriate service response when a problem occurs?
  • Is lower reliability a good trade-off for shorter repair time?
  • When does it make sense to time-phase capacity increases in the data center?

Product Descriptions

Pull-out Battery TrayNpowerDistribution Panel
Pull-out Battery Tray
Pull-out Battery Tray
Pull-out Battery Tray
Pull-out Battery Tray
Pull-out Battery Tray
Pull-out Battery Tray
Figure 1. Liebert Npower physical layout.

The Liebert Npower (Figure 1) is a costeffective 40 kW UPS system designed for highreliability with internal construction that supports easy maintenance and repair.It is available in a broad range of fixed design points. Npower arrives on sitefully tested and assembled, ready for installation by an electrical contractor orfacility electrician and startup by a field engineer. Its human interface is a convenientgraphic display with mimic panel that shows system power flow. Npower uses a singlestring of 40 batteries housed in a separate rack with a second, independent stringadded for redundant systems. Batteries that operate up to 38 minutes at the maximumdischarge rate are available.

The second product is a highly modular UPS system designed to allow a manager toadd power capacity in small increments. This system is composed of a power distributionrack that accepts up to four 10 kW power UPS modules, plus an additional modulefor redundant operation. Each power module functions as a full UPS when pluggedinto the power distribution rack and each requires a string of 32 batteries foreach 6-minute increment of operation. The 32-battery increments are arranged ingroups of four modules, each containing eight batteries.

UPS ModuleHighly Modular UPS System
Input Isolation Transformer
Input, Output and Bypass Circuit Breakers
UPS Controls and User Interface
Distribution Panel Boards
UPS ModuleBatteries
UPS ModuleBatteries
UPS ModuleBatteries
UPS ModuleBatteries
System Logic
Comm Module
Battery Breaker
Static Bypass
Figure 2. The highly modular UPS system physical layout.

The configuration shown in Figure 2 has 384 batteries for up to 18 minutes of fullload operation.

Both systems provide conditioned power to the load and present a friendly face tothe utility. In addition, they have an automatic bypass switch to provide unconditionedpower to the load if the UPS fails or goes offline.

Analyzing Availability and Reliability

Availability is defined as the percentage of time power is supplied to the criticalload. Two factors determine it: system reliability – number of operating hours asystem runs between critical load power failures – and down time – the period fromthe moment of critical power failure to its restoration. For the purposes of thispaper, the term availability is used only in reference to critical load power. Factorsother than power can impact availability and should be considered when calculatingthe expected availability of business critical systems.

Experienced IT customers have become more demanding. Exposing critical applicationsto unreliable utility power has become unacceptable – the cost is simply too high.This is the driving force behind the popularity of redundant UPS systems that havereduced dependency on bypass power and moved critical load power availability topreviously unattainable levels.

Npower has been on the market since 2001 and has demonstrated extraordinarily highreliability. Critical load power availability of Npower systems exceeds 99.999 percent,which corresponds to a system mean-timebetween- failure (MTBF) of 1,000,000 hours.

Highly modular systems are a more recent entry into the market. Assessing theirreliability relative to Npower requires an examination of the key differences betweenthe two systems, particularly in the areas of component count and battery configuration.

Component Count

One of the basic truths of reliability analysis is that the more parts a devicehas, the more frequently that device will fail. A modular product has more partsbecause it has multiple modules, and it exhibits a proportionally higher predictedfailure rate than a fixed-capacity UPS.

This reduction in module-level reliability can partially be compensated for by increaseddesign margins and by using the modules in redundant configurations.

ElementComponentComponent Count
  NpowerHighly Modular UPS System
1 x 40 kW1 x 10 kW4 x 10 kW
Rect/Chg/InvAll of above1238152
Table 1. Each Npower UPS power converter has 12 power components.The highly modular UPS system has 38 per module. A 4-module version of this systemhas 152 (4 x 38) power components. Consequently, it is likely to experience 12 timesmore component failure than Npower.

Careful circuit selection is a must to prevent further decreases in module-levelreliability. However, other highly modular systems are designed with circuits witha high component count, so their module-level reliability is compromised even further.When the module component count is multiplied by the number of modules, the negativeimpact on reliability is dramatic. Table 1 compares the power component count ofboth systems.

Figure 3 illustrates the direct effect component count has on reliability. Notethe dramatic reliability reduction as modules are added. This reduction is due inpart to modularity, but stems primarily from circuit design. Ultimately, the availabilityof critical load power is limited by this reliability decrease.

Figure 3Figure3. The reliability of Npower's power converter is assigned the value 1.0.Relative reliability calculations reflect higher power module component counts only.Similar effects of higher battery and connector counts are not considered.

Battery Configuration

Batteries are the data center’s last stand if the power goes out. They’re also theleast reliable UPS element because poorly maintained batteries are a major causeof critical load power loss. Well-designed, easy to maintain UPS battery configurationsare vital to long-term data center reliability.

Npower uses only 40 batteries with 80 heavy-duty bolted connectors for maximum connectionintegrity. A common highly modular UPS system has 32 batteries with 64 slip-on connectorsper power module. In a 40 kW redundant configuration of this system, 6 minutes ofoperation requires 128 batteries. Extending the operating time to 18 minutes triplesthe number to 384 batteries and 768 slip-on connectors – and doesn’t bode well forbattery reliability. Because of this high battery count, a faulty battery that addslife-shortening stress to the rest of the string is more likely to go undetected.


The IT manager has to decide the appropriate level of protection for the data center’smost critical applications. Non-redundant systems typically provide excellent powerprotection at a reasonable cost. If a system fails, it automatically transfers tobypass power. However, the load then depends on unreliable, unconditioned utilitypower. A redundant system provides an additional layer of protection and dramaticallyreduces the risk of power interruption at the critical load. Redundant UPS systemsare the most reliable and provide the highest degree of protection and availability.

The redundant Npower 1+1 achieves the highest level of protection possible by connectingtwo complete, identical, fixedcapacity systems in parallel, each operating at 50percent of rated capacity. This 100 percent redundancy combined with low systemstress makes Npower extremely reliable. If one system fails, the other automaticallyassumes the full load until the failed system is repaired and returned to service.Figure 4 demonstrates this design, where simplicity correlates with high reliability.

Figure 4Figure4. Complete redundancy is achieved in the Npower 1+1 configuration, wheresimplicity correlates with high reliability. Note the minimal common connectionpoints between the power modules.

Since Npower 1+1 is designed with two independent battery strings, if a fault occurson one battery bus, the other system supports the full load because its batterystring is unaffected.

In the alternative UPS system, the power module is redundant but the battery busis not. A fault on this bus might cause all UPS modules to go offline, leaving theload – and the data center – on bypass.

Figure 5 shows the block diagram for this system. The predicted availability ofhighly modular systems is limited by the relatively high failure rate of the powermodules and by the common bus battery configuration. The service philosophy forthis system is based on the rapid swap out of failed modules, or of the entire systemif necessary, to restore system availability.

Figure 5Figure5. Highly modular system redundancy model. Note the many common connectionpoints between the power modules.

Dealing with Downtime

No manager wants to deal with downtime, but it can happen. If a failure occurs,there is a risk that a person unfamiliar with the system may make a mistake thatcauses the critical load to drop.

Npower is backed by a global service organization that offers a variety of serviceplans, including four-hour on-site response by factory-trained customer engineersfamiliar with preventive and remedial procedures. It features an optional auto-dialmodem to automatically dispatch a service engineer to the site and notify appropriatedata center personnel as well as other advanced monitoring capabilities that supportcomprehensive, network-based or stand-alone monitoring.

For systems without this resource, data center personnel have to do much of theirown service work and quickly replace failed modules. This means already busy ITpersonnel must be trained to provide 24 hours a day, seven days a week coverage.If a technician can’t be present or a spare battery fails, or if the problem isnot in a swappable module, the uncertainty and confusion may delay the arrival ofa qualified technician. Subsequent diagnosis and repair will take longer than ifa qualified technician had been called immediately. In addition, the data centermanager has the added burden of ensuring technicians are both trained and practicedin the maintenance of a UPS system.

Also, consider that many non-UPS factors contribute to system failures, includingsystem grounding and neutral issues, loose or corroded cable terminations, and transformeror breaker failures. Proper installation, combined with an effective preventivemaintenance program can reduce or eliminate these potential problems

System Availability

Availability is determined by reliability and downtime. It is stated as a fractionor a percentage and usually equated to average annual downtime. Npower 1+1, at anavailability of 99.999998 percent, is predicted to average less than 0.1 minuteper year of downtime. A highly modular system with redundancy is predicted to experienceroughly 40 times that amount.

 Predicted AvailabilityPredicted Average Annual Downtime
Npower 1+1 (Redundant)0.99999998< 0.1 min
Highly Modular System (Redundant)0.999993204 min
Npower (Nonredundant)0.9999651018 min
Table 2. Predicted Reliability for configurations in Figures4 and 5.

Cost of Ownership

Cost of ownership is a complex issue. Data center construction, power equipment,heat removal equipment, and equipment racks are major items to consider; primarycost drivers are floor space and power consumption. Managers need to decide if time-phasedcapacity increases make sense given initial data center requirements and predictedgrowth rate.

Floor Space

IT managers know that servers, not power supplies, generate revenue. Finished, fully-supporteddata center space is prime real estate and very expensive – from $200 to $450 persquare foot, depending on the size of the center. Server rack power density industry-wideis predicted to grow from its current range of 1 kW – 3 kW per rack to 10 kW – 15kW per rack within a few years. This rapid rise in power density is driving deploymentof modular, rack-oriented heat management systems.

Npower gives the data center manager the power to choose where to place the UPS:on the data center floor or in a less expensive equipment room. It requires onlyfront access for installation and maintenance so can be located against a wall andbetween already-installed equipment.

The comparative product can be installed only within the data center. The distributionpanel is not modular, so the UPS must be installed near the server racks where floorspace is most valuable. Typical configurations consume 10-20 percent of the availabledata center rack space. This increase in floor space directly drives constructioncost. And locating UPS equipment in the data center requires 10–15 percent higherair conditioning system capacity to handle the added efficiency load.

Figure 6Figure6. Comparison of Npower and highly modular UPS system prices for typicalconfigurations.

Power Equipment Cost

The price of a non-redundant 40 kW Npower is about half that of a 40 kW highly modularsystem, including spare parts and startup. The higher comparative predicted reliability,combined with the increased data center construction cost brought about by locatingthe UPS in the data center tips the scale even more in favor of Npower. Figure 6demonstrates typical product prices.

Spare Parts

Liebert warranties and service contracts provide spare or replacement parts as needed,with no additional cost associated with Npower spares. Without these warrantiesand service contracts, the customer must invest in spare parts so the person on-sitethat is responsible for maintenance can swap power modules, battery modules, and/orcontrol modules when failures occur.


An often overlooked cost is the time and labor of UPS pretest and installation.Each Npower is factory-tested by trained test technicians who test all Npower subassembliesand internal modules. System-level tests include full load and overload testing.Every system is operated for 24 hours, then shipped as a complete UPS configuredfor each customer’s specific needs. Installation is simple: a contractor puts pre-assembledsections in place and interconnects them.

If the system were to arrive in individually packaged cartons, a field engineermust assemble the product before an electrical expert installs it, and the engineerperforms startup.

Cost Analysis

Figure 7 demonstrate the cost differences in two typical data center situations.In the first, total capital expense is dominated by data center construction, andpurchase and installation of power equipment. Because significant additional moneyis needed to expand air conditioner capacity and racks, incremental additions shouldbe considered for these items. Full-capacity installations make sense when incrementalcost is low. They avoid expansion-related construction errors and save money byeliminating multiple electrician visits and avoid endless capital acquisition forms,justifications, and re-approvals.

Figure 7Figure7. Initially, the power and environmental systems are matched to serverrack loads (see the first bar of each pair). The systems subsequently expand asthe load grows to the per-rack capacity limit.

The second chart shows the difference in capital required to build a minimum capacitydata center, and that needed to build out to maximum capacity, is substantial foreach item. The incremental approach may be wise in such cases. It’s up to the datacenter manager to decide whether to do the complete installation up front or waituntil the load demands it.

Cost of Downtime

The true cost of data center downtime is almost immeasurable. Interruptions of criticalapplications cause financial losses for data center customers and financial penaltiesfor data centers. System failures, even those that don’t result in downtime, erodecustomer confidence.

Choosing the right UPS – one that’s backed by a strong, proven service organization– is crucial to long-term data center reliability.


Liebert’s Npower is the right UPS choice for small to midsized data centers. Designedfor reliability and backed by a strong service organization, Npower meets the needsof data centers dedicated to preventing problems before they occur.

Npower Findings:

  • Proven design provides superb reliability and availability
  • Global service organization offers topcaliber support, maintenance, and repair
  • Lower cost of ownership
  • Full capacity available at installation
  • Complete redundancy, including batteries

Inherently, highly modular UPS systems focus on quick repair of plug-in moduleswith higher failure rates. They allow high reliability to be traded away for modularitybased on the premise that shorter repair time compensates for more frequent systemfailures.

Highly Modular UPS System Findings:

  • Circuit design, not modularity, is the primary reason for its power section componentcount increase
  • Non-redundant battery defeats system redundancy
  • Repair time is fast only if spare parts and trained customer personnel are available24/7
  • Risk is high because of technicians unfamiliar with power technology
  • Cost of ownership is higher because of increased component count, floor spaceutilization, heat management, installation and maintenance considerations, and potentialcost of downtime

Liebert’s design philosophy has paid off with an enviable field reliability record.Npower is the clear choice for small to midsized data centers.

Contact your Liebert service representative for more details.


25 окт 2010Datacenter Infrastructure in Sofiaповече

20 яну 2010Liebert XDFN Panduit Editionповече

21 дек 2009Liebert CRVповече

White papers

01 май 2011Application Considerations for Cooling Small Computer and...повече

01 апр 2011Longevity of Key Components in Uninterruptible Power Systemsповече