The trend away from the traditional data center towards the Hyperscale cloud Data center, which has started to gain momentum in the last decade, has also had an impact on the annual power consumption of ICT. The following article describes how the power consumption of data centers has changed since then, how it will change and how the cloud contributes to it. Furthermore, current research on energy efficiency in the cloud is presented and possible problem solutions are discussed.
When I talk to administrators and developers about the cloud, we often think of the same positive aspects: better utilization of existing hardware, high availability, flexible scaling. But I‘m not just a developer, I‘m a scientist, too, and I‘m concerned with the energy consumption of software and hardware. A topic that fits perfectly into our time, in which students go out on the streets and demonstrate each and every day, for a better awareness of global warming and environmental pollution. As Jeff Barr, Chief Evangelist for AWS, described in his blog „Cloud Computing, Server Utilization, & the Environment“, the way to the cloud can reduce energy consumption by up to 84%. This results from “[…] 77% fewer servers required (i.e. cloud re- quires only 23% of the number of servers required for the same workloads) by 71% more efficient servers […]”(1). So, can we see the cloud as the savior of energy consumption in modern data centers? Well, that would be going too far. Although Jeff Barr‘s statement in 2015 is absolutely correct, other factors play a role in the change of the data center landscape. In the following I will briefly explain which changes will occur and why we should deal with them more intensively, on the one hand, of course, when thinking of our children and future generations and on the other hand in the sense of meeting the requirements and challenges of administrators or developers.
Development of the cloud and consequences
In November 2018, Cisco released its Cloud Global Index2. An overview of the changes in data centers over the last few years, including a forecast up to 2021, showing that the number of Hyperscale data centers will grow by 13% annually between 2016 and 2021.
Figure 1: Global hyperscale data center growth(2)
In contrast, the number of classic data centers will decline slightly. Accordingly, the overall energy consumption of data centers should also be reduced, based on the facts already learned. However, this is a misconception, because the already mentioned features such as increased accessibility or flexibility also increase the demand for cloud services. Data in the cloud, check, apps in the cloud, check. A life without the cloud is no longer imaginable and dependence continues to rise. As a fact, already between 2016 and 2018 the traffic to cloud data centers increased annually by 27%.
So we‘re in a vicious circle. With every improvement of the cloud (and also its energy consumption), the demand for cloud services increases immensely, which leads to additional data center infrastructure and so on. One major reason for the increasing data flows is online video streaming, as the Shift Project shows in its study „Unsustainable Use of Online Video“. According to this study, 60% of the total online data flow is generated by online video streaming. Further triggers are increasing computing-intensive applications in the field of artificial intelligence and the confirmation of the proof of work of Bitcoin and similar blockchain systems. All these factors result in an annual increase in worldwide electricity consumption, which has an impact on the climate on the one hand, and on administrators and developers in terms of price on the other. More and more researchers are becoming aware of this problem, especially from a climate point of view, as they try to find or at least draw attention to solutions in the field of Green ICT (information and communication technology). For example, Anders Andrae of Huawei Technologies (Sweden R&D Center) discussed the global power consumption of data centers in his lecture „Predictions on the way to 2030 of internet‘s electricity use“(3).
Figure 2: Cloud data center traffic growth(2)
In 2018, this amounted to around 211 TWh, representing 1% of global electricity consumption. In October 2017, he predicted that by 2025, data centers would use 1200 TWh, and in the worst case even more than 3000 TWh of energy(4). This prediction was slightly adjusted over time and extended to 2030 but remained in many people‘s minds. In this respect Ralph Hinteman et al(5) are concerned with a comparison of different prediction models, in order to be finally able to set up their own model.
Figure 3: Energy consumption of servers and data centers worldwide – forecasts to 2030(5)
Current measures and achievements
Even if these models differ from each other, they all have one thing in common: the power consumption of data centers will continue to rise. The Borderstep Institute is currently working on the study „Energy-efficient Cloud Computing Technologies and Policies for an Eco-friendly Cloud Market“(6), which focuses on sensitization and solution finding. Workshops and expert interviews have already taken place for this purpose. There are also projects that deal intensively with the efficiency standards of computer systems. ISO Standard 23544 (Information Technology – Data Centers – Application Platform Energy Effectiveness (APEE)) is currently under development(7). In addition, ETSI – EN 303 470 has been an active standard since March 2019 and deals with „Energy Efficiency measurement methodology and metrics for servers“(8).
Furthermore, research projects are dedicated to topics that directly affect data centers, for example, to the energy efficiency of software, which implicitly influences the energy consumption of computer systems. As shown in figure 4, IT as a whole accounts for approximately 45% of data center power consumption.
Figure 4: Power distribution in data Centers(9)
Current research with focus on software
The load driver is the software running on the machines. In order to be able to draw more attention to this subject area, the awarding of a blue angel for energy-efficient software is currently under development in Germany(10).
All in all, it can be said, that the demand for energy efficiency in the ICT sector is increasing. There are an increasing number of projects that are put in the public’s focus. I was able to experience this myself, because when you are dealing with a practical topic such as the energy efficiency of container systems, for example, you are often rejected at conferences either because you are working too practically in terms of a research topic or because you are working on a too exotic topic for practitioners. Actually, it should be clear that if I design my containers to be energy efficient, the huge mass of containers can also save a lot of energy. Then why isn‘t this interesting for administrators and developers? The reason is that end users, for example a user of SAAS, FAAS or PAAS, do not notice the increasing energy consumption of the data centers, because with an in- creasing number of servers, more and more users are added to the cloud. The costs remain evenly distributed and the user continues to pay according to the duration of use of his services. But services are not the same and have to be considered in different ways, because there are acute fluctuations here. A small website in a container, with few accesses per day, does not consume as much power as a REST-API including a database with millions of entries accessed every minute. In my opinion, we have to rethink here. If we know how much energy our services consume and if we optimize them accordingly, the overall power consumption of data centers will decrease on one hand, and price models in the cloud area can be adjusted with even more flexibility on the other. A calculation based on CPU and RAM is not enough, because other factors such as network IO also play a role. It‘s not that we can‘t control these things. But both admins and developers should be encouraged to be energy efficient. I‘m currently working on recommendations for energy-efficient container handling. The administration and orchestration as well as the development of applications with containers play a role. The goal is to save energy every time a container is started and stopped, every time it is scaled, and even beyond that over the entire lifetime of the container. Especially in scaling, the consideration of the power consumption of the services and not only their CPU utilization offers various possibilities to save energy. As I mentioned before, this results from the sheer mass of containers used in cloud data centers. As can be seen in table 1, the number of workloads and compute instances per server, such as VMs and containers, is rising with an annual rate of 22%.
Table 1: Workload and compute instances in cloud data centers(2)
As already mentioned, even the slightest effect on these calculation instances can make a big difference.
In conclusion, the cloud has made the lives of administrators, developers and end users easier and more adapt- able than ever. As development continues in the PAAS, SAAS, FAAS and similar areas, there will be more and more data centers around the world that will consume more and more power. In terms of a sustainable and environmentally conscious way of life, administrators and developers also have to face new problems and have to take responsibility and thus, solutions have to be found accordingly. Current research clearly points out the problems and offers first solutions, which give administrators and developers not only a clear conscience, but also the opportunity to offer new and more flexible pricing models in the cloud.
If you are interested in supporting me in my research in the field of energy efficient use of containers, I would be glad if you could participate in the following short survey on your usage of containers and, in particular, of Docker: http://www.incubator-solutions.com/survey
1. Jeff Barr, Cloud Computing, Server Utilization, & the Environment, AWS News Blog, June 2015, https:// aws.amazon.com/de/blogs/aws/cloud-computing-server-utilization-the-environment/, last accessed 11/09/2019
2. Cisco Global Cloud Index: Forecast and Methodology, 2016–2021 White Paper, November 2018, https://www.cisco.com/c/en/us/solutions/collateral/ service-provider/global-cloud-index-gci/white-paper-c11-738085.html, last accessed 11/09/2019
3. Anders Andrae, Predictions on the way to 2030 of internet’s electricity use, The Lost Decade? Planning the Future, University Copenhagen, February 2019
4. Anders Andrae, Total Consumer Power Consumption Forecast, Nordic Digital Business Summit, October 2017
5. Ralph Hintemann, Simon Hinterholzer, Energy Consumption of Data Centers Worldwide – How will the Internet become Green?, ICT4S, Lappeenranta,
6. European Commission, Study on energy-efficient cloud computing technologies and policies for an eco-friendly cloud market, 2019, https://ec.europa. eu/digital-single-market/en/news/study-energy-efficient-cloud-computing-technologies-and-policies-eco-friendly-cloud-market, last accessed 11/09/2019
7. ISO/IEC CD 23544, Information Technology — Data Centres — Application Platform Energy Effectiveness (APEE), 2019, https://www.iso.org/standard/76000. html, last accessed 11/09/2019
8. ETSI – EN 303 470, Environmental Engineering (EE); Energy Efficiency measurement methodology and metrics for servers, https://standards.globalspec.
com/std/13240725/EN%20303%20470, last accessed 11/09/2019
9. G. Smpokos et al, On the Energy Consumption Fore- casting of Data Centers Based on Weather Conditions: Remote Sensing and Machine Learning Approach, April 2018, arXiv:1804.01754
10. Green Software Engineering, Set of criteria for sustainable software, https://green-software-engineer- ing.de/en/kriterienkatalog-v01/introduction.html, last accessed 11/09/2019
He is a researcher at Trier University of Applied Sciences,
Environmental Campus Birkenfeld and a Ph.D. student at
Trier University in the field of energy efficiency of cloud
technologies. He works as lecturer at the environmental
campus Birkenfeld and is head of technology at the
venture capital donor capacura. He graduated in applied
computer sciences and mathematics (M.Ed.) in 2015.