Responding to market needs, server technology has rapidly advanced, delivering high-density, highly reliable computing without the massive harmonics issues inherent with older power supplies. One unintended consequence of this advance is a shift in load power factor characteristics in data centers. Prior to the server technology advancements, data center power factors ranged from 0.6 to 0.9 lagging. Now, it's not unusual to see power factors ranging from 0.9 leading to 0.9 lagging.

Uninterruptible power supplies (UPSs) of older designs cannot deliver their full rated power to these new loads and must be derated or retrofitted. Therefore, data center designers and operators must be vigilant to avoid problems such as insufficient capacity, nuisance alarms, and even downtime due to overloads. Fortunately, new UPS designs can avoid these issues while delivering more real power to the load.

Power supply evolution

As noted above, servers historically had power factors ranging from 0.6 to 0.9 lagging. These designs had many advantages but produced high input current harmonics in addition to the poor power factor. Eventually, regulatory pressure led to standards such as the EN/IEC 61000-3-2 standard in the European Union, forcing power supply manufacturers to address this issue.

The server industry response was to add large input filter capacitors to the power supply. Although these were optimized for full load, servers at the time routinely had a single power supply typically loaded at 40% to 50% of capacity. Nonetheless, they were effective in reducing harmonics while load power factors remained predominantly lagging (inductive).

As server technology became more powerful, servers assumed more of the mission-critical computing tasks. In order to achieve the higher reliability required for these applications, servers were equipped with hot-swappable redundant power supplies. This reduced typical power supply loading from approximately 50% down to approximately 25%, while the filter capacitor was still typically optimized for 100% loading.

In modern “blade” server systems, multiple server cards are supported by a common set of redundant power supplies. This reduces the number of components, especially fans, and thus increases reliability. However, this further reduces the loading of power supplies, while leaving the filtering capacitors sized for the highest load scenario should one power supply fail. Additionally, many blade chassis are not fully populated or used. With these changes, server power factors now fall into a range from 0.9 leading (capacitive) to 0.9 lagging (inductive), depending on design and loading at any particular moment.

The challenge

Most large UPS module designs were not ready for this development. To understand why not, you must consider the key UPS subsystems and how they operate. A double conversion UPS has two major subsystems: a DC subsystem (rectifier) and an AC subsystem (the inverter). Each has a limit in terms of how much power it can deliver. The DC system limit is defined in kW, and the AC system limit is defined in kVA. Overloading either of these overloads the UPS.

Older inverter designs have large output filters based largely on capacitance. If the load power factor is lagging (inductive), the reactance of the load and the filter offset each other — and the UPS can deliver all or most of its rated kW to the inductive load. However, if the load power factor is capacitive (leading), the capacitive reactance of the filter and the load become additive. This causes the UPS to use more of its kVA capacity to overcome this capacitive reactance.

Figure 1 (click here to see Fig. 1) is a power output diagram for a typical UPS optimized for 0.7 to 0.9 lagging power factor loads. This shows the constraints on UPS power delivery at various load power factors. The vertical axis “A” represents active or true power (watts), while the horizontal axis “B” represents reactive power (VARs or kVARS). The lengths of the vectors originating at point “AB” equal kVA. The vertical height of any point on a vector indicates the kW. The horizontal distance from axis “A” to any point on a vector represents the kVAR of a load at that point. The relationship of the vectors and axis is defined by the geometry of a right triangle:

kVA = √[(kW)2 + (kVAR)2]

The angle between the vector and the vertical axis “A” is θ, which is the angle of deflection of the current wave form from the voltage waveform. The cosine of θ is the power factor. The constraints on UPS power delivery are represented by lines “C,” “D,” and “E.

  • C” is the kW limit of the DC subsystem of the UPS (rectifier and battery).

  • D” represents the kVA limit of the UPS inverter. It is offset because of the capacitive nature of the output filter, causing it to favor lagging (inductive) loads.

  • E” represents the range of real loads, or practical VA

You can see the effects of these constraints in Table 1 (click here to see Table 1). The vectors extend from the center point until they meet a constraint. The light blue shaded area represents the range of loads that can be supported. On the leading side, the UPS inverter kVA limit is the binding constraint in this example. On the lagging side, the constraint is the rectifier kW until you reach a region (darker blue area) where the load does not demand all of the kVA or kW that the UPS can supply.

Some UPS manufacturers recognized this shift in power factor and developed designs that are compatible with leading power factor loads. Output filters in these designs incorporate a balance of capacitive and inductive characteristics that give the UPS a symmetrical output, as seen in Fig. 2 (click here to see Fig. 2).

Note that the UPS kVA limit (“B”) is now concurrent with the practical VA limit shown in Fig. 1. Additionally, the latest designs for North America incorporate a kW/kVA rating based on delivery of full kVA and kW at a 0.9 power factor. The range of supportable loads is shown by the shaded area in Fig. 2.

Note that a larger percentage of targeted loads can be supported, and that de-rating is not necessary anywhere within this range. Table 2 (click here to see Table 2) shows the kVA and kW that the UPS can deliver at various load power factors, and provides a comparison with the UPS described in Fig. 1 and Table 1.

The new UPS designs with symmetrical output characteristics offer significant advantages to the data center designer and operator:

  • The UPS can be specified based on its kW/kVA rating, subject only to normal practice of 80% maximum loading for safety factor.

  • Derating is not required for modern loads nor for any likely to develop over the next several years.

In effect, the operator will know that dynamic changes of load level or power factor within the specified kW/kVA rating of the UPS will not cause overloading and the resultant nuisance alarms or downtime.

Heller is product manager for 3-phase UPS systems at Chloride North America in Libertyville, Ill. He can be reached at chuck.heller@chloridepower.com.

Load Power Factor Delivered kVA % of Unit kVA Rating Delivered kW % of Unit kW Rating Limit
0.7 lagging 500 100% 350kW 87.5% Load demand
0.8 lagging 500 100% 400kW 100% Rectifier kW
0.9 lagging 444 88.8% 400kW 100% Rectifier kW
0.95 lagging 421 84.2% 400kW 100% Rectifier kW
1.0 (unity) 400 80% 400kW 100% Rectifier kW
0.95 leading 358 71.6% 340kW 85% Inverter kVA
0.9 leading 342 68.4% 308kW 77% Inverter kVA
0.8 leading 320 64% 256kW 64% Inverter kVA
0.7 leading 309 61.8% 216kW 54% Inverter kVA

Table 1. UPS power delivery at load power factors from 0.7 lagging to 0.7 leading for 500kVA/400kW UPS (0.8 PF) optimized for lagging power factor loads.

Modern UPS with Symmetrical Output 0.9 PF Legacy Design UPS 0.8 PF (From Table 1)
Load power factor Delivered kVA % of rated kVA Delivered kW % of Unit kW rating Limit Delivered kVA % of Unit kVA rating Delivered kW % of Unit kVA rating
0.7 lagging 500 100% 351kW 78% Load demand & inverter kVA 500 100% 350kW 87.5%
0.8 lagging 500 100% 396kW 88 % Load demand & inverter kVA 500 100% 400kW 100%
0.9 lagging 450 100% 450kW 100% Rectifier kW 444 88.8% 400kW 100%
0.95 lagging 474 94.8% 450kW 100% Rectifier kW 421 84.2% 400kW 100%
1.0 (unity) 450 90% 450kW 100% Rectifier kW 400 80% 400kW 100%
0.95 leading 474 94.8% 450kW 100% Rectifier kW 358 71.6% 340kW 85%
0.9 leading 500 100% 450kW 100% Rectifier kW 342 68.4% 308kW 77%
0.8 leading 500 100% 396kW 88% Load demand & inverter kVA 320 64% 256kW 64%
0.7 leading 500 100% 351kW 78% Load demand & inverter kVA 309 61.8% 216kW 54%

Table 2. Comparison of 500kVA/450kW UPS (0.9 PF) optimized for 0.9 leading to 0.9 lagging loads vs. legacy design (500kVA/400kW UPS 0.8 PF) optimized for 0.9 to 0.7 lagging loads.