- The increasing digitalization of the economy and society is leading to a dynamic increase in the amount of data that is processed and stored in data centers. Some scientific studies assume that this growth will also lead to a significant increase in the energy consumption of data centers worldwide. This article deals with the questions of whether and to what extent the energy consumption of data centers have increased in the past and what future developments can be expected. In addition, future challenges to sustainable data center operation will be analyzed. The article deals in particular with the topics of improving the energy efficiency of IT hardware and software, the use of regenerative electricity for the operation of data centers, and the use of waste heat from data centers. Index Terms- Data centers, energy consumption, waste heat, regenerative electricity, Bitcoin mining, energy efficiency, IT hard- and software Against this background, the present contribution addresses the following questions: • How did the energy consumption of data centers and their individual components develop in Germany between 2010 and 2017? • What are the results of different studies on the development of the energy consumption of data centers in a global comparison? • Which current trends influence the energy consumption of data centers? • How can the energy consumption of data centers develop up to the year 2030? What forecasts are likely? • What are the particular challenges for energy-efficient operation of data centers in the future? II. METHODOLOGY
Video streaming, social media, big data, Bitcoin, artificial
intelligence, and digitalization of business processes and
production flows – these and other trends are leading to more and
more data being stored and processed in data centers. Data
center capacity is growing dynamically. The IT company Cisco
assumes in its analyses that the worldwide computing
capacities of data centers measured in workloads and computer
instances will more th
Whether this significant growth in computing and storage capacity will also be accompanied by an increase in energy consumption can hardly be predicted from today's perspective. However, most available analyses assume a more or less significant increase in energy consumption [2]–[14]. Yet the range of forecasts is very wide. While Andrae/Edler, for example, predict a 15-fold increase in the energy consumption of data centers worldwide between 2010 and 2030 [3], Shehabi et al. found that the energy consumption of data centers in the USA has stabilized in recent years due to increasing energy efficiency [4]. For more than 10 years, the authors themselves have been analyzing the development of the energy consumption of data centers with the help of a comprehensive structural model of data centers in Germany [9], [14]–[25]. For Germany, an increase of 25% in the energy consumption of data centers in the period from 2010 to 2017 has been calculated [27]. Simon Hinterholzer (Author)
This paper deals with the energy requirements of data centers, presenting the findings from various international studies and comparing them with the authors’ own calculations. The calculations are carried out with the help of an extensive structural model of the data center landscape in Germany which was developed at the Borderstep Institute and is updated annually. In this model data centers are defined as all enclosed spatial units such as server cabinets, server rooms, building parts, or entire buildings in which at least three physical servers are installed. The energy requirements of stand-alone servers are also calculated. The development of the data center capacities is calculated in particular on the basis of the IT equipment (servers, storage, networks) in the data centers. The various size classes of the data centers in Germany are described in the model in terms of their equipment with different server types, storage systems, and network infrastructures. The model also takes into account the age structure of the servers and the energy requirements of the different server types in different operating states. In addition, the data center infrastructures such as • • • • air conditioning, power supply, UPS, etc. are modeled for different size and redundancy classes.
The following sources in particular were used for the calculations: •
Study “Entwicklung des IKT-bedingten Strombedarfs in Deutschland“ (Development of ICT-related electricity consumption in Germany) by Fraunhofer IZM and Borderstep on behalf of the Federal Ministry of Economics and Energy [26], Current and publicly available results of studies on the development of the data center market from various analyses [16], [17], [28]–[35], Data from the market research institute Techconsult on market development for servers, storage, and network components (eanalyzer) [36]–[38], Data from the market research institutes IDC and EITO on market development for servers, storage, and networks in Germany and Europe [39], [40], Scientific literature and manufacturer information on the development of energy consumption for servers, storage, and network products, and for advanced efficiency technologies for data centers.
A detailed description of the model has already been provided in various publications. [9], [15], [25], [26], [41].
III. DEVELOPMENT OF THE ENERGY CONSUMPTION OF
DATA CENTERS IN GERMANY
Using the structural model of data centers presented above,
it was calculated that despite a significant improvement in the
energy efficiency of data centers, their energy consumption in
Germany increased by 25% between 2010 and 2017 (Figure 1).
The increase was mainly due to an increase in the number and
performance of IT components in the data centers. In 2017,
around 2.37 million servers were in operation in Germany,
18% more than in 2010. The number of data centers equipped
with IT hardware for data storage and network technology also
increased significantly – the number of hard disks in Germany
more than doubled between 2010 and 2017. As a result of this
increase, the power consumption of IT components in data
centers rose from 5.8 billion kWh in 2010 to 7.9 bi
Significant further improvements in the efficiency of data center infrastructures can be expected in the future. Newly built larger data centers today often realize PUE values of 1.3 and lower [27], [33].3
Growth in the data center market is driven primarily by the
cloud market (Figure 2). Large international cloud providers in
particular are currently building up resources in Germany and
are thus meeting the requirements of German companies to
enable the storage and processing of their data in Germany.
International cloud companies often use the support of
colocation providers, thereby driving the colocation market at the
same time. The share of colocation data centers in the IT area
of all data centers in Germany again increased significant
From a regional perspective, the greater Frankfurt area and thus the federal state of Hesse in particular are benefiting from 1 2
The Power Usage Effectiveness (PUE) value indicates the ratio of the total energy consumption of a year to the annual energy consumption of the IT department of the data center.
The stand-alone servers, which are normally operated without their own air conditioning, are not included in the calculation of these values. Taking stand-alone servers into account, the average PUE value in Germany improved from 1.82 in 2010 to 1.58 in 2017. 3
The PUE is an indicator often used to assess the energy efficiency of data centre infrastructures. The advantage of PUE is that it is easy to determine. However, the interdependencies between the energy demand of IT and the energy demand of the entire data center are very complex. If, for example, the energy consumption of IT is reduced, the PUE can worsen, even though the total energy consumption of the data center decreases significantly. the growth of the data center market. This applies in particular to the colocation data center segment [34]. In 2017, there was a total IT area of 550,000 m2 in the data centers in Hesse, 250,000 m2 of which was in colocation data centers. As a result, the capacities in the colocation data centers – measured by the IT space in Hesse between 2010 and 2017 – have increased by 60% (Figure 3). In terms of the available IT connection resources, capacities even increased by 100%. [34].
Despite the dynamic development in the data center market in Germany, it is possible that it will become less important in international comparison as a data center location. Especially in the rapidly growing segment of hyperscale data centers, which are very large, significantly more capacity is being built worldwide, also elsewhere in Europe, for example, the Netherlands, Ireland and Scandinavia. [33], [34].
Operators of hyperscale data centers shy away from Germany as a business location primarily because of the high electricity costs and the sometimes lengthy approval processes. The increasing shortage of skilled personnel for data centers in Germany is also becoming an increasingly disadvantageous location factor [33]. Rapidly growing applications such as Bitcoin mining [6] cannot be operated in Germany at breakeven due to the high electricity prices.
EUROPE AND WORLDWIDE
Several scientific studies examine the energy consumption of data centers. In the following, the results of selected and well-known studies are briefly presented and compared.
The results of a 2015 paper by Andrae/Edler [3] have achieved a relatively high level of attention in the discussion about the energy consumption of data centers [13], [14], [42]– [44]. This is certainly due to the fact that Andrae/Edler forecast a strong increase in energy consumption. In the "expected" scenario, according to their calculations, the energy consumption of data centers worldwide will increase by a factor of 15 from approx. 200 billion kWh/a in 2010 to almost 3,000 billion kWh/a by 2030. Even in the "best" scenario, an increase by a factor of almost 6 to 1,337 billion kWh/a is calculated. In the "worst" scenario, the energy consumption of data centers will increase by a factor of 40 to 7,933 billion kWh by 2030 [3].
At the beginning of 2019, Andrae published an update of its
calculations and forecasts, which deviates significantly f
A publication by Belkhir and Elmeligi [14] also assumes a
very significant increase in the energy requirements of data
centers. Based on a calculation by Vereecken et al. [46], they
assume a worldwide energy demand for data centres of 275
billion kWh/a in 2009 and anticipate an annual growth rate of
10% by 2020. This growth rate was determined by the market
research company Technavio [47]. This results in an energy
demand of 659 billion kWh/a for data centers wor
Malmodin/Lunden calculate worldwide energy
consumption of data centers to be 240 bi
According to a Borderstep Institute estimate, worldwide
energy consumption of server data centers increased by about
30% to 287 billion kWh between 2010 and 2015 [9]. This
increase accelerated once again in the last two years. A current
TEMPRO project estimate conc
The different assessments of the development of the energy consumption of data centers are also reflected in regional analyses. For the USA, Shehabi et al. calculated that the energy consumption of data centers has hardly increased since 2010; they calculate energy consumption of 70 billion kWh in 2014, compared to approx. 67 billion kWh in 2010. The small increase in the USA is due to a general improvement in the energy efficiency of IT systems and infrastructure with lower PUE values. In addition, there has been a significant shift in computing power to particularly efficient hyperscale data centers with PUE values below 1.2 [4], [10].
If one compares the development of the energy consumption of the data centers in the USA calculated by Shehabi et al. with the development in Germany calculated by Borderstep, it is striking that energy consumption in Germany seems to increase significantly more strongly, although the increases in the server inventories are comparable. The basic structure of Shehabi et al.'s model is comparable to that of the Borderstep Institute. Therefore, it was possible to carry out a detailed comparison of the different model parameters. The differences in the developments can essentially be attributed to two factors. On the one hand, there are hardly any hyperscale data centers in Germany – therefore, computing power has hardly shifted to this particularly efficient type of data center. Secondly, the assumptions in the models differ with regard to the average development of the maximum power consumption of a server at full load. While Shehabi et al. assume that maximum power consumption is constant [4], the Borderstep model assumes an increase in maximum power consumption due to a significant increase in the average amount of RAMs and multiprocessor systems. These assumptions are also confirmed for Europe by other studies [7], [11]. Whether the differences in the model assumptions are justified by the real differences in the USA and Europe or Germany is yet to be verified by further investigations.
The results of various studies on the energy consumption of
data centers in Europe are relatively similar. The Ecodesign
Preparatory Study on Enterprise Servers and Data Equipment
determines energy consumption of 78 billion kWh fo
In summary, the various studies on the development of the energy consumption of data centers do not provide a uniform picture. While some studies, e.g. Andrae/Edler, assume a very strong increase, other studies assume low to moderate growth in the last ten years.
However, the energy efficiency of the data center infrastructure and thus the PUE values have improved significantly in recent years, and the share of IT components in the energy consumption of the data centers has thus increased.
CONSUMPTION OF DATA CENTERS DEVELOP IN THE
FUTURE?
As the results presented in the previous section show, there are great uncertainties in determining the energy consumption of data centers worldwide even for the current point in time. Forecasts beyond 2020 are even more difficult due to the unclear development of technologies and the extent to which data centers and the services they provide will be used in the future [4], [10]. In particular, the possible end of the efficiency advances made to date, and thus the end of Moore's Law, could lead to a significant increase in energy consumption in the future. Andrae/Edler, for example, have modeled an end to Moore's Law in their calculations by 2022 and thus also an end to the significant increases in data center efficiency [3]. This is a major reason for the very significant increase in energy consumption in their scenarios. It can be discussed if and when Moores Law will end. It is clear that technical development is approaching the physical limits of conventional CMOS silicon technology. The structures currently consist of only a few atomic layers. Leading companies such as Intel and AMD expect Moore's law to come to an end around 2023 [50]. However, alternative materials and 3D architectures will presumably contribute to further increases in performance and reliability despite the small structures. An abrupt end to improvements in the energy efficiency of computing operations is unlikely.
Assuming a continuation of Moore's Law, the Andrae/Edler model yields much more moderate increases. Another reason for the significant increases in energy consumption in the model is the assumption that the performance of data centers is determined by the IT traffic between the user and the data center. According to Cisco forecasts, this increases by 23% annually [3], [51]. If the assumption of increasing data center performance is modified in such a way that the number of workloads and compute instances is chosen as the measure, the calculated increase in energy consumption in the model would be lower. According to Cisco, the number of workloads and compute instances is currently growing at an average annual rate of 18.6% [1].
To illustrate the differences between the different analyses, figure 5 shows various forecasts of the energy consumption of data centers worldwide through 2030. Studies have been selected which allow the presentation of the development up to the year 2030. Beside the investigations of Andrae/Edler [3], Andrae [45] and Belkhir/Elmeligi [14] a “best case” is presented. This development would occur using the Andrae/Edler model if Moore's Law continued and moderate growth rates in data center performance were assumed. Furthermore, according to Borderstep's calculations, the trend in the development of the worldwide energy consumption of data centers between 2010 and 2017 will continue through 2030.
A comparison of the various forecasts shows that the
possible future development of the energy consumption of data
centers has a wide range. In the “best case” the energy
consumption of data centers can remain largely constant. If Moore's
Law ends and the performance of the data centers increases
significantly, annual energy consumption may increase to
almost 3,000 billion kWh/a (And
However, the analysis of the development makes one thing clear: energy-efficient operation of data centers will continue to be of great importance in the future. However, the challenges are changing. This will be discussed in the following section.
HARDWARE AND SOFTWARE, USE OF WASTE HEAT AND
RENEWABLE ENERGIES
As shown above, the PUE values of data centers are improving continuously. As a result, the share of the data center infrastructure in the energy consumption of data centers is becoming smaller. In the future, efforts to improve the energy efficiency of data centers will have to focus even more on IT components. Further improvements to IT hardware, energyefficient software, and efficient software deployment models such as virtualization and container technology offer opportunities for optimization. Completely new technologies such as neuromorphic processors or the use of artificial intelligence to improve efficiency in data centers also offer high potential [18].
Even if all possible future improvements in the efficiency of IT components and infrastructure are taken into account: almost everywhere, the electricity used in data centers is converted into heat and then released into the environment – mostly using additional energy for ventilation and cooling. The example of Sweden in particular shows that it is possible to use the waste heat from data centers under appropriate conditions [52], [53]. In view of the increasing energy consumption of data centers, the topic of waste heat utilization will play a central role in the future [18]. Innovative new solutions must be implemented here. This applies to both classic air-cooled and new innovative liquid-cooled IT systems. In Germany and presumably in other countries too, however, the framework conditions must change in order to promote the use of waste heat in data centers. So far, electricity prices in Germany have been so high that the operation of heat pumps that can raise the waste heat level of data centers to a usable level is uneconomical. Often it is economically more favorable to burn oil or gas for heat generation instead of using existing waste heat.
Another future challenge for environmentally friendly data center operation is the supply of electricity generated largely from renewable sources because the supply of continuously available electricity generated from hydropower or biomass is limited. This means that intelligent use of fluctuating wind and solar power in data centers is becoming increasingly important. Here, too, there are promising technological approaches [18], [54], [55].
This paper discusses the development of the energy consumption of data centers, and the results of various studies are presented. Although the various studies presented assume greater or smaller increases in energy consumption of data centers in recent years, estimates of both the absolute amount of energy consumed and the increases in energy consumption differ significantly.
The forecasts for the development of future energy consumption of data centers differ even further. The article briefly presents and discusses possible scenarios for the energy consumption of data centers worldwide through 2030, ranging from keeping energy consumption constant to an increase by a factor of 40.
All the uncertainty and variation of the forecasts notwithstanding, a further significant increase in the energy consumption of data centers seems likely. Improving energy efficiency will therefore continue to be of great importance. The focus here will be more on improving the energy efficiency of IT components in the future, as significant improvements in infrastructure such as cooling and uninterruptible power supply have already been achieved in the past. Measures that do not directly affect the energy efficiency of data centers, such as the use of waste heat and operation with (fluctuating) regenerative energy, will also become increasingly important in the future.
With a view to the future, it can be assumed that the focus of sustainable data center operation will shift from the sole consideration of energy requirements to other sustainability categories. Initial approaches, e.g. for the consideration of the 17 Sustainable Development Goals of the UN, already exist today [56].
This contribution is a result of the project TEMPRO – Total
Energy Management for professional data centers, which is
funded by the German Federal Ministry of Economics and
Energy. We would like to thank all project partners and especially
project manager Alexandra Pehlken for their constructive and
excellent cooperation.
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
http://www.independent.co.uk/environment/global-warmingdata-centres-to-consume-three-times-as-much-energy-in-nextdecade-experts-warn-a6830086.html. [Accessed:
28-May2018]
N. Jones, “How to stop data centres from gobbling up the
world’s electricity - The energy-efficiency drive at the
information factories that serve us Facebook, Goog
K. Fichter and R. Hintemann, “Beyond Energy: Material
Stocks in Data Centers, Taking Resource Efficiency into
account in Green IT Strategies for Data Centers,” J. Ind. Ecol.,
no. im Er
R. Hintemann, “Consolidation, Colocation, Virtualization, and
Cloud Computing – The Impact of the Changing Structure of
Data Centers on Total Electricity Demand,” in ICT
Innovations for Sustainability. Advances in Intelligent
Systems and Computing, L. M. Hilty and B. Aebischer, Eds.
Berlin, Heidelberg:
R. Hintemann, K. Fichter, and D. Schlitt, “Adaptive
computing and server virtualization in German data centers
Potentials for increasing energy efficiency today an in 2020,”
in Proceedings of the 28th Conference on Environmental
Informatics - Informatics for Environmental Protection,
Sustainable Development and Ri