White Paper - Infrastructure in an energy-efficient data centre Verena Schwalfenberg Dipl.-Ing. Markus Schmidt
Copyright © 2008
All rights reserved.
Rittal GmbH & Co. KG Auf dem Stützelberg
D-35745 Herborn
Phone +49(0)2772 / 505-0 Fax +49(0)2772/505-2319
www.rittal.de www.rimatrix5.de
V10
Infrastructure in an energy-efficient data centre
Contents
List of figures ........................................................................................................................................... 3
Abbreviations........................................................................................................................................... 4
Executive Summary................................................................................................................................. 5
Introduction.............................................................................................................................................. 6
Energy-related problems in data centres ................................................................................................ 7
Infrastructure ....................................................................................................................................... 7
Development of electricity costs and consumption ............................................................................. 7
Reasons for the increasing energy consumption.............................................................................. 10
Concepts and options for optimization .................................................................................................. 13
Power supply and distribution ........................................................................................................... 13
Cooling .............................................................................................................................................. 14
Room cooling................................................................................................................................ 14
LCP............................................................................................................................................... 16
Air/water heat exchanger.............................................................................................................. 16
Free cooling .................................................................................................................................. 17
Outlook .................................................................................................................................................. 19
2
Infrastructure in an energy-efficient data centre
List of figures Figure 1 Development of electricity costs (industrial) in Germany.......................................................... 7
Figure 2 Procurement costs for new servers compared to the costs for power & cooling...................... 8
Figure 3 Increase in server energy density ............................................................................................. 9
Figure 4 Relationship between operating costs over 3 years and server procurement costs ................ 9
Figure 5 Watts per US$ 1,000 for 1U servers ....................................................................................... 10
Figure 6 RimatriX5 efficiency monitor ................................................................................................... 11
Figure 7 Efficiency of a 40kW UPS ....................................................................................................... 12
Figure 8 RimatriX5 “Pay-as-you-grow” concept .................................................................................... 13
Figure 9 Flexible scalability of PMC 200 ............................................................................................... 14
Figure 10 Raised-floor cooling with cold / hot-aisle design ................................................................... 15
Figure 11 Cold-aisle containment.......................................................................................................... 15
Figure 12 Cold-aisle containment & LCP Inline cooling ........................................................................ 16
Figure 13 Average annual temperatures in Germany ........................................................................... 17
3
Infrastructure in an energy-efficient data centre
Abbreviations
CO2 - Carbon dioxide
COP - Coefficient of performance
CPU - Central processing unit
IDC - International Data Corporation
IT - Information technology
kW - Kilowatt
kWh - Kilowatt hour
LCP - Liquid Cooling Package
MTTR - Mean time to repair
PDM - Power Distribution Module
PDR - Power Distribution Rack
PMC - Power Modular Concept
DC - Data centre
SI-EER - Site infrastructure energy ratio
UPS - Uninterruptible power supply
4
Infrastructure in an energy-efficient data centre
Executive Summary
This white paper, focusing on the infrastructure in energy-efficient data centres, provides an overview
of the basic energy problems in existing data centres. It explains which components in an
infrastructure place the biggest load on the energy budget.
Innovative solutions are needed as additional power consumption leads to increased waste heat,
which in turn has be cooled using energy-intensive methods. The ever-rising electricity costs also
make this a top priority. It is predicted that by 2010, electricity and cooling costs will account for 50%
of the total costs of a data centre1.
There are a number of reasons for this rise in energy consumption, including the growing demands on
computing power and the continuous increase in the packaging density in the server area. At the
same time, there is also huge potential savings offered by e.g. IT chillers, which often remains
untapped.
RimatriX5 is one solution that helps improve energy efficiency. It is a modular concept from a single
source that meets the demands for an energy-efficient data centre infrastructure.
5
1 Source: International Data Corporation (IDC) 2006
Infrastructure in an energy-efficient data centre
Introduction
Energy efficiency is one of the key terms used in IT today in conjunction with data centres. There are
numerous reasons for this development. The increasing demands on computing power, rising server
packaging densities, more waste heat, CO2 emissions and, above all, the huge rise in energy costs
automatically push for energy-efficient solutions to help cut electricity consumption.
A few years ago, the day-to-day energy costs for data centre operation played a relatively insignificant
role. More emphasis was placed on procurement costs for the IT hardware, such as servers, storage
systems and network technology. Over time, this technology has become increasingly powerful, which
has caused power consumption to rise accordingly. It is now clear that this level of power consumption
leads to considerable costs and problems. That is why manufacturers of IT hardware have
increasingly been integrating energy-saving mechanisms into their products. These include flexible
clock rates, lower operating voltages and sections of processor chips that can be deactivated as
necessary.
So far, little attention has been paid to the data centre infrastructures used to provide and distribute
the energy required by the IT hardware and to dissipate waste heat. In a worst case scenario, this
energy consumption is higher than the consumption of the IT hardware itself. Therefore, innovative
concepts and solutions are also required in the infrastructure to boost energy-efficient operation.
6
Infrastructure in an energy-efficient data centre
Energy-related problems in data centres
Infrastructure
The infrastructure of a data centre must solve two basic problems in relation to energy:
Reliable provision of energy
Dissipation of the heat that is generated
These are both key cost factors in day-to-day operation and have a major impact on the efficiency of a
data centre.
In simplified terms, the efficiency of a data centre can be defined as follows:
DC
ITDC nconsumptiopower
nconsumptiopowerefficiency__
=
ytechnonetworkstorageservernconsumptiopower IT log__ ++=
)(__ chillercoolerPDMUSVnconsumptiopowernconsumptiopower ITDC ++++=
The efficiency of a data centre depends on the infrastructure technology required for the
uninterruptible power supply (UPS), power distribution (PDM), coolers and water chillers. The lower
the efficiency levels of these individual components, the poorer the overall efficiency of the data
centre.
If the infrastructure components exhibit a poor level of efficiency, this has a double impact on the
overall system – the energy costs rise and the server's power consumption increases.
Development of electricity costs and consumption
Figure 1 Development of electricity costs (industrial) in Germany2
Electricity costs for the commercial sector have been increasing steadily since 2000, and the trend
looks set to continue as a result of the worldwide demand for energy.
7
2 Source: Energieagentur NRW (Energy Agency NRW)
Infrastructure in an energy-efficient data centre
Furthermore, servers today consume more power than they did a few years ago despite state-of-the-
art architectures, which is causing the absolute consumption values for the infrastructure to skyrocket.
Figure 2 Procurement costs for new servers compared to the costs for power & cooling3 It is not surprising, therefore, that the power and cooling costs are assuming a bigger share of the
overall costs of a data centre. In Figure 2, the IDC forecasts that energy costs will account for around
40% of the total IT budget in 2010, while other experts predict that this figure may even rise to 50%.4
In addition to high energy consumption and the associated costs, increased energy consumption is
giving rise to a further problem in that existing data centres can no longer handle the power
requirements and waste heat.
3 Source: International Data Corporation (IDC) 2006
8
4 Source: (“CEO Guide to Green Computing”, businessweek.com 2007/05).
Infrastructure in an energy-efficient data centre
Figure 3 Increase in server energy density5
The energy density in data centres increased more than 10-fold between 1994 and 2004 alone. This
rise places extreme demands on the infrastructure of the power supply (UPS), power distribution and
cooling concept. This is because every watt of power supplied must also be removed from the data
centre in the form of waste heat.
Figure 4 Relationship between operating costs over 3 years and server procurement costs6 According to calculations from the Uptime Institute, the server operating costs skyrocket compared to
procurement costs when users continue to work with old concepts and structures. In the best case
(projection B)7 the increase is still considerable, but projection A8 is realistic. However, the old
infrastructures must be modernised to achieve these values.
5 Source: Datacom Equipment Power Trends and Cooling Applications, 2005
9
6 Source: Data Center Energy Efficiency and Productivity, Uptime Institute 7 State-of-the-art
Infrastructure in an energy-efficient data centre
Further evidence of an obsolete infrastructure is its SI-EER value. The site infrastructure energy
efficiency ratio is the ratio between the energy supplied and the energy that reaches the IT hardware.
It is the reciprocal value of EfficiencyDC.
According to the calculations of the Uptime Institute, this value came to an average of 2.5 in 2007.
This means that of the 2.5 watts supplied, only 1 watt reaches the IT hardware. In a data centre
covering 2,800 square meters, reducing the SI-ERR to 2.0 would cut energy costs by around
US$ 1 million a year.
Reasons for the increasing energy consumption
To be able to implement cost-saving measures and therefore run a data centre more efficiently, it is
important to identify the reasons behind the rising energy consumption.
The first and most obvious reason for this is the fact that the performance in the server area is increasing all the time. Moore’s Law9 states that the complexity of the chip circuitry doubles every 24 months. This doubling of integration density goes hand in hand with a significant increase in computing power. As a result of the growing demands, more and more servers are being operated in each data centre. Therefore, the total power consumption of data centres rises even though the value for “watt per computing power” falls significantly. This phenomenon becomes clear if you consider how many server watts could be purchased over the
years for US$ 1,000.
Figure 5 Watts per US$ 1,000 for 1U servers10
From 2000 to 2006 alone, the power per US$ 1,000 increased more than 13-fold. Even in best case
scenarios, the possible developments (projections A and B) predict a rise to 157 watt/US$ 1,000.
One reason for the considerable increase in recent years is the blade server. Its extremely compact
design means that a very high power density can be generated in the rack. A rack equipped with
several blade centres can easily result in power consumption greater than 20kW, which then has to be
cooled. With an SI-EER of 2.5, the data centre has an input power of 50kW for just one rack!
8 Best Practice
10
9 A law devised by Gordon Moore in 1965 that states that that the number of transistors that can be inexpensively placed on an integrated circuit increases exponentially, doubling approximately every two years 10 Source: Data Center Energy Efficiency and Productivity, Uptime Institute
Infrastructure in an energy-efficient data centre
Many servers are bought to host one or two special services. These applications often only place a
load of between 10 and 20 percent on the server, which means that 80% of the CPU’s computing time
is spent on idle operation. This would not be a major problem if power consumption was scaled linear
to the computing load. Unfortunately, this is not the case. A server with only a 20% load needs around
75% of the energy that it needs in productive operation with a 70 – 80% load. This problem cannot be
solved by the infrastructure, but it places a tremendous load on it.
Figure 6 RimatriX5 efficiency monitor
The RimatriX5 efficiency monitor illustrates the problem with infrastructures that are oversized or not
operated to their full capacity. The zoom view shows that the efficiency of the data centre (bottom
graph, blue curve) drops rapidly to less than 10% when the IT load (top graph, green curve) is far
below the normal load.
11
During the planning of a data centre, the infrastructure is often designed to cater for maximum server
configuration. In other words, the capacities of the UPS and cooler in particular far outstrip the
requirements in the first few years of operation. In addition to causing very high procurement costs, at
worst the devices also operate at less than 30% of their maximum load and, therefore, with a
significantly lower level of efficiency.
Infrastructure in an energy-efficient data centre
Efficiency 40kW UPS
88.0%
89.0%
90.0%
91.0%
92.0%
93.0%
94.0%
95.0%
15% 30% 52% 76% 102%
Load
Effic
ienc
y
Wirkungsgrad
Figure 7 Efficiency of a 40kW UPS
When operated to 90% of its capacity, a 40kW UPS, for example, has an efficiency level of approx.
94.5%. However, if it is operated at just 25% of its capacity, the efficiency level drops to 91%.
With a power supply of 100kW, a drop in the efficiency level of 3.5 percentage points increases power
losses by 3.5kW. Over a year, this amounts to 30,660kWh or €3,066 (based on electricity costs of
€0.10 per kWh). This loss of €3,066 is caused purely by poor sizing, which also has to be paid for at
the investment stage (bigger UPS = higher procurement costs). Similar calculations can also be done
for cooling.
Solutions in this area must, therefore, ensure that components deliver optimum efficiency throughout
the service life of a data centre.
Cost savings Runtime kWh Saving [€]
1 year 30,660 3,066 5 years 153,300 15,330
10 years 306,600 30,660 Potential savings from improved UPS efficiency11
12
11 Reduced power loss of 3.5kW
Infrastructure in an energy-efficient data centre
Concepts and options for optimization
Power supply and distribution
As mentioned in the previous section, two main factors impact on the power supply of a data centre –
efficiency and sizing.
Sizing means choosing the right-sized components for the system, and it should be based on the
demand and the final expansion stage of the data centre. When data centres were planned in the
past, only the maximum configuration was factored into the equation “to be on the safe side”. This
configuration is often only achieved in the final years of a data centre – if at all.
To enable infrastructure components, such as UPS and power distribution, to operate at maximum
efficiency during the initial phase, they must be configured slightly smaller, with the option of
expansion at a later date. This balancing act can be realised with modular systems such as RimatriX5
from Rittal.
1
2
3
Figure 8 RimatriX5 “Pay-as-you-grow” concept12 Legend 1 – Current power consumption 2 – Scalable, demand-oriented and gradual adaptation of capacity (RimatriX5) 3 – Overcapacity configured with conventional DC technology x-axis: Service life in years y-axis: Installed infrastructure in % / room capacity Figure 8 demonstrates the pay-as-you-grow concept from RimatriX5. The components for UPS, power
distribution and cooling grow as the data centre expands.
This modular concept can be illustrated using the example of Rittal PMC 200 UPS.
13
12 Source: Rittal GmbH
Infrastructure in an energy-efficient data centre
Figure 9 Flexible scalability of PMC 200
The PMC 200 can be operated using three 20kW modules as an (n+1)-redundant 40kW UPS. If new
servers demand more power, there is no need to buy new or additional UPSs. The power output can
be increased simply by inserting another module whilst the system is operational. These modules are
available in several classes ranging from 8 to 40kW, enabling flexible adaptation to the power
consumption required. The modular n+1 redundant design offers energy-saving benefits for both
upgrades and operation. A second UPS is not required in the event of failure – an additional module
with less power loss is sufficient. The modular concept also boosts availability. During service
scenarios, a faulty module can be replaced in just a few minutes. This results in a very low MTTR
(mean time to repair). The state-of-the-art system design without a transformer cuts down on the use
of copper in production and makes for above-average efficiency of more than 95%.
The same modular concept with all its energy-saving benefits can also be realised for power
distribution. Rittal offers such a system with its Power Distribution Rack (PDR) and Power Distribution
Modules (PDM).
Cooling
Room cooling The second biggest factor that impacts on the efficiencyDC value is data centre cooling. Even with an
average SI-ERR of 2.0, a data centre with a 100kW IT load creates almost 200kW heat losses that
have to be cooled.
The traditional method is room cooling, whereby cold air is blown through the room – or better,
through a raised floor into the DC. In older data centres, this arrangement can cool a heat load of
between 1 and 2kW per rack; in newer, better planned DCs, this value increases to 5kW. However,
this solution demands very accurate implementation of the cold-aisle concept. Figure 10 shows this
type of cold-aisle scenario.
14
Infrastructure in an energy-efficient data centre
Figure 10 Raised-floor cooling with cold / hot-aisle design13 It is essential that hotspots or heat leakages, which can cause the cold and warm air to mix, are
avoided. This reduces the efficiency of the cooling output considerably and threatens the reliable
operation of the servers at these points.
Figure 11 Cold-aisle containment14
The next optimisation step is cold-aisle containment, whereby cold air is directed using additional
ceilings, walls and doors. This method is some 10-20% more efficient than traditional cold-aisle
cooling without containment.
13 Source: Rittal GmbH
15
14 Source: Rittal GmbH
Infrastructure in an energy-efficient data centre
Figure 12 Cold-aisle containment & LCP Inline cooling15
LCP If the cold air supplied via the raised floor is no longer sufficient, LCP Inline – i.e. cooling devices
inserted between the racks – is the next step to optimise the cooling efficiency. LCP Inline devices
from Rittal are the ideal solution, offering a high level of efficiency and excellent reliability. With a
Liquid Cooling Package (LCP), the air is cooled via an air/water heat exchanger, enabling very high
cooling outputs.
However, if several fully-equipped blade centres are installed in the racks, the cooling load in the
enclosure is too high for traditional or optimised room cooling concepts. In this case, rack-based
cooling is virtually compulsory. The encapsulated rack is cooled and does not affect the temperature in
the server room. The LCP Modular or LCP Plus system can be used to cool the rack. The cooling
output of these systems, which are also based on air/water heat exchangers, can be adapted and
extended step-by-step via cooling modules, in line with the RimatriX5 pay-as-you-grow concept. The
maximum cooling outputs for LCP Modular and LCP Plus are 20kW and 30kW respectively.
In very critical applications, a redundant design can be realised with the LCP system with 2 LCPs
being used to cool one rack.
Air/water heat exchanger In cooling solutions using air/water heat exchangers, the cooling water must be made available with a
specific inlet temperature and the heated water from the return flow must be cooled again. IT chillers
are usually used for this purpose. The power of the IT chiller must be adapted to the waste heat that is
generated. Rittal offers a finely-graded range with cooling outputs between 15kW and 462kW. These
power levels in themselves indicate that cooling requires an enormous amount of energy, coupled with
substantial costs and CO2 emissions.
16
15 Source: Rittal GmbH
Infrastructure in an energy-efficient data centre
Annual average temperatures in Germany In degrees Celsius
long-term average
linear trend
Thermometer values on the rise
Figure 13 Average annual temperatures in Germany16 Free cooling Free cooling offers good savings potential for IT chillers. A free cooling system uses ambient air to
cool the heated cooling water. This method can always be used to generate cold water when outside
temperatures are low.
Free cooling aims to cut operating costs and reduce greenhouse gases. Hot air is extracted without
the use of a compressor, which reduces the total electrical energy required by the IT chiller, thus
making it much more efficient.
In standard applications, the IT chiller can be operated in free cooling mode with an outside
temperature of around 10 degrees Celsius. On average, the temperature in Germany is below 10
degrees for almost 50% of the year. As air-conditioning in a data centre usually runs 24/7, the savings
potential is also almost 50%. In standard cooling mode, a 462kW chiller requires an input power of
96.25kW.
Cooling efficiency Q=462kW; coefficient of performance17: COP=4.8
Required input power for the IT chiller: kWkWCOPQP 25.96
8.4462
===
When the IT chiller is operated in free cooling mode, the required input power drops to around 10kW.
Based on an electricity price of 10 cent/kWh, this results in potential savings of almost €38,000 every
year!
16 Source: DWD / FAZ graphic
17
17 Thermal efficiency of heat pumps, identical with the term coefficent of performance
Infrastructure in an energy-efficient data centre
Sample calculation for IT chiller costs with and without free cooling
Pow
er c
onsu
mpt
ion
of
chill
er w
ith c
ompr
esso
r [k
W]
Ope
ratin
g ho
urs
per
yeea
r
Cos
ts [€
]
Pow
er c
onsu
mpt
ion
of
chill
er w
ithou
t co
mpr
esso
r [kW
]
Ope
ratin
g ho
urs
per
yeea
r
Cos
ts [€
]
Tota
l cos
ts p
er y
ear [
€]
Without free cooling
96 8,760 84,096 10 0 0 84,096
With 30% free cooling
96 6,132 2,452 10 2,628 2,628 61,495
With 50% free cooling
96 4,380 42,048 10 4,380 4,380 46,428
With 70% free cooling
96 2,628 25,228 10 6,132 6,132 31,36018
Compared to costs without free cooling, a system that uses 50% free cooling saves 376,680 kWh a
year, which is equivalent to €37,668/year.
This system works by feeding the water via an air/water heat exchanger to the outside air where it is
cooled by temperature balancing with the ambient temperature (<10°C). This process can be
enhanced by fans. This method is efficient because it dispenses with the need for a compressor which
would consume power.
18 The calculation is based on an electricity price of 10 cent/kWh
18
Outlook
The importance of energy efficiency for data centre operators will continue to grow. The need for
action will also increase in view of the costs, energy availability and environmental factors. RimatriX5
from Rittal is an innovative and modular concept from a single source that meets the infrastructural
requirements for a modern, energy-efficient data centre. The modular design of RimatriX5 offers a
wide variety of options for expanding and modernising existing data centres, regardless of whether the
operator wants to install a complete new cooling system or just a single rack.
The next optimisation steps for the RimatriX5 range are already in development and will ensure that
existing and new installations will continue to be energy-efficient in the future, too.