siemens.

com/datacenters

Distributed Redundancy in
data center applications
How power supply configurations increase reliability and
optimize investment and operational cost

White Paper | November 2014

The Information Communication and Technology sector is moving at a rapid pace. It
transforms our life vision and generates an increasing demand for centralized pro-
cessing power and stored information. As a consequence, Data Center facilities grow in
both number and size.

Over two-thirds of global electricity production is based on the consumption of fossil

fuels. Therefore, as major and increasing energy consumers, Data Centers are inextri-
cably linked with global energy challenges and climate change.

Generally, the optimal efficiency point (or 'sweet spot') of critical power distribution
components occurs at the rated power capacity. However, as a result of redundancy
requirements, critical power systems in Data Centers are operated below their rated
capacity. Depending on the conceptual design and especially in dual cord server con-
figurations, the critical power system can operate well below 50% during normal oper-
ation. Needless to say that this not an optimal situation from neither the efficiency nor
the investment point of view.

The conceptual design of the critical power path itself is a major contributor to the
maximum operational load of the system during normal operation. The Distributed
Redundancy concept is one of the designs that is implemented to increase normal load
without sacrificing availability and reliability. It therefore has a positive effect on both
investment and operational cost.

This white paper elaborates on the design implications of the Distributed Redundant
power distribution concept for dual cord server applications.

Written by David Meulenbroeks

Siemens Center of Competence Data Centers, The Netherlands.

A white paper issued by: Siemens. 1

© Siemens AG 2014. All rights reserved
 White paper | Distributed Redundancy in data center applications | November 2014

Contents

3 Distributed Redundancy: the basics

3 Distributed Redundancy over two supply

systems
4 Distributed Redundancy over three supply
modules
5 Distributed Redundancy over four supply
modules
6 General characteristics
7 Impact of Distributed Redundancy on data
center facilities
8 Acronyms and variables

A white paper issued by: Siemens. 2

© Siemens AG 2014. All rights reserved
 White paper | Distributed Redundancy in data center applications | November 2014

Distributed Redundancy: the basics

Distributed Redundancy over two supply systems
Distributed Redundancy in its simplest form is probably the best known design concept for critical power supply systems
in data centers. It is also referred to as the ‘True A-B’ design. Figure 1 gives a simplified overview of this design. It shows
normal operation as well as the power balance shift in case of a B-side main supply fault.
The figure presents the two main electrical operational modes for a data center: Normal operation and Emergency oper-
ation.

300 300 600 0

PSM PSM PSM PSM

A B A B

100 100
50 50 100 0
Load Load

100 100
50 50 100 0
Load Load

100 100
50 50 100 0
Load Load

100 100
50 50 100 0
Load Load

100 100
50 50 100 0
Load Load

100 100
50 50 100 0
Load Load

Normal Operation Emergency Operation

Figure 1 Distributed Redundancy over two supply systems (True A-B design)

In this example, the total ICT load [PICT] is presented as the mathematical sum of 6 individual loads of 100 1; the black fig-
ures in the picture. These loads are dual cord and it is assumed that the load is equally distributed over both cords, or 50 -
50; the green values in the picture.

In the Distributed Redundancy nomenclature, the True A-B design as shown in figure 1 is defined as Distributed Redun-
dancy over two supply systems. With these two supply systems, the 6 loads are combined in a single Load Group [LG], as
there is only one combination of supply feeders: A-B. In this example the load of this load group equals 600.

In normal operation, this configuration results in a cumulated load per Power Supply Module (or PSM) of 6 × 50 = 300.
When in emergency operation - with PSM-B off line as shown in figure 1 - the load shifts to the A-side supply. The cumulat-
ed load for PSM-A in that operational mode is therefore equal to the total ICT load [PICT], or 6 × 100 = 600.

In this example, the design is characterized by the following Distributed Redundancy parameters:
Total ICT load [PICT] 600
Normal load per PSM [Pnorm(PSM2)] 300
Emergency load per PSM [Pmax(PSM2)] 600
Installed base PSM [Pinst(PSM2)] 1.200

The index "2" in the variables [E.g. Pmax(PSM2)] indicates to Distributed Redundancy over two supply systems.

1Neither the unit (W, kW or kVA) nor the absolute value of the figures mentioned has significance in this example as the relation between
the figures of the different configurations are relevant, not their absolute value.

A white paper issued by: Siemens. 3

© Siemens AG 2014. All rights reserved
 White paper | Distributed Redundancy in data center applications | November 2014

When considering the ICT-load exclusively (thus excluding power requirements for cooling or other auxiliary systems), the
nominal power rating of the PSM is to be equal or larger than the Emergency load on each PSM: Pmax(PSM2). In the example
above, Pmax(PSM2) equals PICT and is 600.

From these results, the following general relations can be derived. These are exemplary for Distributed Redundancy over
two supply systems.

The normal load relative to the emergency load per PSM is:

Normal load per PSM Pnorm (PSM2 ) 300

= = = 50%
Emergency Load per PSM Pmax (PSM2 ) 600

The installed base 2 Pinst relative to PICT equals:

Total installed base PSM Pinst (PSM2 ) 1200

= = = 200%
Total ICT load PICT 600

Distributed Redundancy over three supply modules

When an additional Power Supply Module is introduced in the design of figure 1, and the load is equally distributed over
all three of the PSM's, the configuration of figure 2 is obtained. Following the nomenclature introduced above, this de-
sign is referred to as Distributed Redundancy over three supply systems.

Again, for this example, the ICT load [PICT] equals 600. The loads are distributed over all combinations of PSM's. This
results in a total of 3 Load Groups [LG] A-B, A-C and B-C each with a normal load of 200.
200 200 200 300 0 300
PSM PSM PSM PSM PSM PSM
A B C A B C

100 100
50 Load 50 100 Load 0

100 100
50 Load 50 50 Load 50

100 100
50 Load 50 0 Load 100

100 100
50 Load 50 100 Load 0

100 100
50 Load 50 50 Load 50

100 100
50 Load 50 0 Load 100

Normal Operation Emergency Operation

Figure 2 Distributed Redundancy over three supply systems

Analogous as described for Distributed Redundancy over two supply systems previously, in this example Distributed Redun-
dancy over three supply systems is characterized by the following parameters:

Total ICT load [PICT] 600

Normal load per PSM [Pnorm(PSM3)] 200
Emergency load per PSM [Pmax(PSM3)] 300
Installed base PSM [Pinst(PSM3)] 900

2The installed base Pinst(PSMn) refers to the nominal output of the PSM. This excludes any redundancy inside the PSM as for example N+1
configured modular UPS systems.

A white paper issued by: Siemens. 4

© Siemens AG 2014. All rights reserved
 White paper | Distributed Redundancy in data center applications | November 2014

The general relations for Distributed Redundancy over three supply systems are therefore:

The normal load relative to the emergency load per PSM:

Normal load per PSM Pnorm (PSM3 ) 200

= = = 66%
Emergency Load per PSM Pmax (PSM3 ) 300

Installed base relative to PICT:

Total installed base PSM Pinst (PSM3 ) 900

= = = 150%
Total ICT load PICT 600

Distributed Redundancy over four supply modules

When the three PSM design of figure 2 is expanded with an additional module, the four PSM configuration of figure 3 is
obtained: Distributed Redundancy over four supply systems.

The example is based on a total ICT load PICT of 600. And again, the loads are distributed over all combinations of PSM's: A-
B, A-C, A-D, B-C, B-D and C-D. This results in a total of 6 Load Groups with a load of 100 each.

150 150 150 150 200 0 200 200

PSM PSM PSM PSM PSM PSM PSM PSM
A B C D A B C D

100 100
50 Load 50 100 Load 0

100 100
50 Load 50 50 Load 50

100 100
50 Load 50 50 Load 50

100 100
50 Load 50 0 Load 100

100 100
50 Load 50 0 Load 100

100 100
50 Load 50 50 Load 50

Normal Operation Emergency Operation

Figure 3 Distributed Redundancy over four supply systems

In this example Distributed Redundancy over four supply systems is characterized by the following parameters:

Total ICT load [PICT] 600

Normal load per PSM [Pnorm(PSM4)] 150
Emergency load per PSM [Pmax(PSM4)] 200
Total installed base PSM [Pinst(PSM4)] 800

A white paper issued by: Siemens. 5

© Siemens AG 2014. All rights reserved
 White paper | Distributed Redundancy in data center applications | November 2014

The generic relations for Distributed Redundancy over four supply systems become:

Normal load relative to the emergency load per PSM:

Normal load per PSM Pnorm (PSM4 ) 150

= = = 75%
Emergency Load per PSM Pmax (PSM4 ) 200

Installed base relative to PICT:

Total installed base PSM Pinst (PSM4 ) 800

= = = 133%
Total ICT load PICT 600

General characteristics
Theoretically, the Distributed Redundancy principle can be expanded infinitely by adding Power Supply Modules. As a re-
sult, both Pnorm(PSM∞)/Pmax(PSM∞) and Pinst(PSM∞)/PICT will approach 100%. In the limit situation - an infinite number of
PSM's - the normal load of the PSM will equal the emergency load. Also, the installed base will equal the total ICT load.

From an installed base and efficiency point of view, this limit situation might look like ideal. However tempting, this limit
obviously represents a hypothetical configuration. With the increasing number of Power Supply Modules, the number of
Load Groups will also become infinite as it equals the sum ∑1+2+3+ .. +(#PSM-1). But besides being infinite in number,
these Load Groups will have a maximum load value Pmax(LG∞) that approaches 0. Furthermore, the distribution system is
going to be infinite as well. Not a very useful configuration.

Table 1 summarizes the main characteristics of the Distributed Redundancy principle with an increasing number of Power
Supply Modules.

Table 1 Main characteristics of Distributed Redundancy with increasing number of PSM's

Number of Number of Load Ratio of normal and emer- Ratio of installed base and Ratio of load group load
supplies Groups gency load on supply load and emergency load on
supply
# PSM [n] # LG Pnorm(PSMn)/Pmax(PSMn) Pinst(PSMn)/PICT Pmax(LGn)/ Pmax(PSMn)
2 1 50 % 200 % 100 %
3 3 67 % 150 % 67 %
4 6 75 % 133 % 50 %
5 10 80 % 125 % 40 %
6 15 83 % 120 % 33 %
7 21 86 % 117 % 28 %
8 28 88 % 114 % 25 %

From this table we can conclude that - depending on the application - the optimum configuration for Distributed Redun-
dancy is somewhere between 4 and 6 Power Supply Modules:

• Below 4 Modules there is relatively much to be gained by adding a PSM as the ratio Pnorm(PSM)/Pmax(PSM) signifi-
cantly increases and the number of Load Groups remains manageable.
• Above 6 Modules the ratio Pnorm(PSM)/Pmax(PSM) flattens and the number of Load Groups becomes impractical to
handle.

A white paper issued by: Siemens. 6

© Siemens AG 2014. All rights reserved
 White paper | Distributed Redundancy in data center applications | November 2014

Impact of Distributed Redundancy on

data center facilities
When selecting optimal power architecture for a facility, data center managers have to consider different factors and un-
derstand the impact of each one on their facility.

When designing the electrical power distribution configuration, the data center designers need to consider both initial and
future load demands. As a result, many data centers are operating at significantly less than 100 percent of design load
much of the time.

Considering the right power distribution load balancing design before a facility is constructed, is a good way to move to
cost efficient data center without compromising on reliability.

When correctly implemented and maintained, the Distributed Redundancy principles meet with the reliability criteria. It can
be applied to the power distribution level of redundancy requested for the data center (N, N+1, 2N, 2(N+1); etc.). Distrib-
uted Redundancy can definitely provide both flexibility on power capacity and cost control optimization without losing data
center reliability.

A white paper issued by: Siemens. 7

© Siemens AG 2014. All rights reserved
 White paper | Distributed Redundancy in data center applications | November 2014

Acronyms and
variables
ICT Information and Communication Technology
LG Load Group
PSM Power Supply Module
PICT Sum of critical load
Pinst Installed power: sum of total installed Power
Supply Modules
Pmax Load in emergency operation mode
Pnorm Load in normal operation mode
UPS Uninterruptible Power Supply

Siemens AG Siemens AG
Building Technologies Division Energy Management Division
International Headquarters Freyeslebenstrasse 1
Gubelstrasse 22 91058 Erlangen
6301 Zug Germany
Switzerland
Tel +41 41 724 24 24
E-mail [email protected]

www.siemens.com/datacenters
All rights reserved. All trademarks used
are owned by Siemens or their respective owners.
© Siemens AG 2014

A white paper issued by: Siemens. 8

© Siemens AG 2014. All rights reserved