Figure 1. ICT benefits factor in 2020 and 2030⁴

Regarding the core (wired) network, power consumption of networking equipment currently mainly depends on the number of active devices. Power consumption of these devices has little dependence on the actual traffic load across all the network segments. Therefore current energy saving approaches mainly focus on exploiting traffic dynamics to deactivate unnecessary resources. They can be classified in three groups: i) dynamic traffic engineering, ii) proxying/virtualization and iii) rate adaptation [8]. However, more fundamental approaches are applicable within the 5G and 6G horizon. They include an energy-efficient architecture as well as design of energy-efficient devices [8] having various modes of operation with quick activation times, reconfigurations upon failures, smart levels of overprovisioning, optical bypassing, etc.

One of the goals of the so-called green computing is to maximize energy efficiency during computing devices lifetime⁵. Approaches to this energy-efficiency include i) data center design, ii) software and deployment optimization, iii) power management, iv) optimized cloud computing, v) supercomputers, vi) edge and fog computing. Regarding data centers, their energy consumption (resulting from IT equipment operation as well as from heating, ventilation and air conditioning) accounts for 1.1–1.5% of the world's total energy use [9]. Regarding software and deployment optimization of the computing machines, algorithmic efficiency, optimized computing resource allocation, machines virtualization and the use of terminal servers are in place to reduce the energy consumption caused both by the computers themselves and their cooling systems. Power management allows a system to automatically turn off or hibernate components in case of user’s inactivity. Cloud computing allows companies for significant reduction of their direct energy consumption by moving their specific on-site applications to the cloud. Supercomputers that may reside in a cloud are getting more and more efficient in terms of performance-per-watt, some of them exceeding 16 GFLOPS/watt. However, as pointed by a Greenpeace study, data centers (the cloud) represent 21% of the electricity consumed by the IT sector, which is about 382 billion kWh a year [10].

III. Communication and Computing Greening Together

It is important to stress that IoT services involve both communication and computing (processing) of information across the network, and thus, key performance indicators including energy-efficiency of communication and computing should be handled jointly. One of the metrics (and main criterion for the algorithm evaluation) that could be proposed is the total number of successfully conveyed (transmitted, processed and received) bits per cost-unit (energy unit or related monetary unit).

The challenges of future massive, yet energy-efficient, communication and computing require new architectural and algorithmic approach to the network design. So far, access to cloud services has been considered as the mainstream enabler for IoT communication and computing. A cloud is commonly understood as a server offering huge memory-capacity for data storing, sharing and analysis, a super-computer with optimized computing algorithms for these data processing, or as a service for end devices connected to IoT. The cloud ecosystem is centralized, thus, requiring transmission of data to long distances. Data centers and cloud servers may be located thousands kilometers away from the user. This, in turn, results in considerable delays, transmission errors and energy cost associated with the use of network resources, although cloud servers may be very effective in terms of the computing energy cost.

The other extreme approach to the IoT communication and computing challenges is fully distributed operation and organization of a network called edge computing. In this case, networking is based on cooperation of networked devices and computing-resource sharing. The basic challenges here are: to assure dynamic adaptation to the changing network traffic and topology, to design context-responsive (context-aware) information processing algorithms, to facilitate dynamic connections and access to the network, to support cooperation and communication/computing resources sharing, to guarantee the quality of service as well as security and privacy of distributed information. The edge devices are not that computationally powerful and energy-optimized as the cloud servers, they require more energy per processed bit, but the communication to long distances and the associated energy cost is avoided.

Fog network architecture has been proposed as more suited for future IoT communication and computing services than centralized (cloud-based) or totally distributed (edge-computing) ones. It can be flexibly organized and managed in response to the variety of 5G/6G envisioned services and their challenges. This architecture is promoted as multi-tier and hierarchical, consisting of the things (edge devices) tier, fog tier and cloud tier [11] and is presented in Figure 3.

In the things tier, IoT end-devices are connected, such as smartphones, tablets, sensors, vehicular communication transceivers, remote-control devices, robots, etc. In the cloud tier, high computing-performance and high memory-capacity servers are placed. In the fog tier, computing machines (microcomputers, personal computers, computer clusters, base-stations, etc.) capable of data processing and analysis, and of communication and cooperation are located. It is important to note that the fog tire can have multiple hierarchical layers, and that communication and cooperation between them is not only vertical, but also horizontal.

In Figure 3, arrows illustrate the data flow in a few use-cases (computations/tasks offloading, multimedia cashing, vehicle-to-vehicle/infrastructure/anything communication, remote control in robotics and data processing). An arrow’s line weight represents the data volume. For example (the orange-arrow case in Figure 3), information content of a large volume, e.g., video, can be acquired by a network of sensors (or cameras), transmitted to a fog node and processed there to extract some specific features. Information of a lower volume, e.g., mentioned features, is transmitted to a data center (a cloud) for storage and some further processing, the result of which, e.g. decision on detected features, might be accessed by interested parties in a larger area covered by a cloud. Another example (red-arrows case in Figure 3) of exploiting the fog architecture is to delegate computational tasks by a  smartphone with small computational power to fog nodes with higher computational capabilities.

In general, the fog architecture offers opportunities for sharing (and thus, uberization) of resources: computational-, communication-, storage- or energy-resources [11]-[13]. It allows for joint (communication- and computing-related) optimization of energy efficiency. Naturally, there is a trade-off between delay, reliability, and energy consumption in the fog, e.g. computing machines in the fog nodes may consume more energy per operation than optimized cloud servers, but provide lower latency [12]. High computational complexity of the processing requests favors processing in a cloud. Delay and power consumption caused by transmitting data through the core network is offset by the high processing speed and computational power efficiency. Conversely, requests which require relatively few operations to process are best served by nearby edge devices or fog nodes.

Finally, the goal of reducing the network energy consumption may not be exactly the goal of a mobile IoT devices that are powered by batteries. In [13], we answer the question whether it is beneficial from the power consumption perspective to perform computations locally by the IoT device or to offload them to the fog. Research prove that usage of fog resources reduces IoT device power consumption mostly in case of short transmission distances, broad transmission bandwidth, and relatively high computational complexity of a task, e.g., more than 300 FLOPs per bit of compressed video stream. Numerical values are valid for the considered, realistic system parameters, but the general trend is expected to be valid for most setups.

IV. Computational Awareness in Links and Networks

On the completion of 3GPP Release 15, the first full set of 5G standards has been defined in 2018, and updated in 2019 and completed with Release 16 in 2020. As in the 4G standard, the Orthogonal Frequency Division Multiplexing (OFDM) has been proposed for 5G systems. Popularity of the OFDM technique results from its known advantages, one of them being adaptation potential to channel conditions. The flexibility in Resource Blocks (RBs) allocation and adaptive modulation and coding schemes (MCSs) may be utilized to maximize the offered transmission rate or minimize the energy consumption. Contrarily to the traditional approach to optimize the transmission parameters in response to channel quality, [14]-[17] look at advanced power-consumption models and optimization of the energy-efficiency (EE) metric defined as the number of successfully transmitted and received bits per Joule. This power consumption modelling must take signal-processing (with necessary computations) power into account. Intelligent utilization of such a model in the link- [14][15] and network- [16][17] energy-efficiency optimization algorithms, can be called computational awareness.

Depending on the link quality (and distance), power consumption of different causes, e.g. RF signal radiation or signal processing, may dominate over each other, and may be worth minimization for overall energy-efficiency. For example, power consumption related to the application of advanced coding schemes and, in particular, computationally-exhaustive decoding, may be higher than transmission power in short links, where the channel quality is good. In Figure 4, trends are shown for the computationally-aware link-adaptation in OFDM. There, one can see that this awareness results in higher EE than the rate-maximizing water-filling algorithm. The exact numbers depend on the specific scenario, some indicative figures can be found in [14][15].

In a more advanced network scenarios, such as relay (decode-and-forward -DF) interference networks, incorporation of the computational awareness in link-adaptation, dynamic cooperation-mode selection and relay-node selection algorithms results in increased energy-efficiency of the network, when compared to standard solutions. Relevant results and comparisons are provided in [16][17]. For example, Figure 5 shows that computational-awareness in energy-efficiency optimization implies that the cooperative mode is selected by some users even though interference is observed in the second time slot of relaying, and additional computations are necessary in the DF operation. This is because of high interference-channel attenuations, and relatively low communication- and computation- energy cost for these users (located at short distance from the relay).

Note that, in future massive communication in IoT, many devices may be transmitting on short links, many of them may serve as relays, and therefore application of computational awareness in such  networks is crucial. The challenges are to develop distributed algorithms and assure this awareness. Moreover, the optimization algorithm itself also consumes energy, that needs to be taken into account.

V. Conclusions

All reports and forecasts for 10 years horizon show that ICT business is, and will be doing very well. However it also faces the existential challenge from climate change. The expectations are that the ICT sector will not only reduce its own carbon footprint, but also deliver significant reductions in CO₂e emissions to other business sectors. The pressure is on researchers, engineers, business and governments to apply innovations, strategies and ambitious policies that could reverse the growth of the ICT sector footprint and enable global emissions reduction.

Over 3 trillion US Dollars is likely to be spent on research and development in the ICT sector up to 2030, indicating huge potential for innovative solutions to the SDGs [4]. Some promising technologies for greening communication and computing in future massive-type of networks are at the maturation stage and have been presented above.

[4] Based on the slideshow by Luis Neves, GeSI Chairman, "#SMARTer2030: ICT - Disruptive Technologies for Sustainable and Better Living” at 5th ITU Green Standards Week, Nassau, The Bahamas.

[5] There are other goals related to reduction of hazardous materials and factory waste, recyclability or biodegradability, which we do not address here.

Acknowledgement

The presented ideas have been worked out within the FAUST project (no. PL-TW/V/3/2018) funded by the National Centre of Research and Development, Poland.

References

1. The Ericsson Mobility Report, June 2020, https://www.ericsson.com/49da93/assets/local/mobility-report/documents/2020/june2020-ericsson-mobility-report.pdf
2. Global e-Sustainability Initiative, The Climate Group, “SMART 2020: Enabling the low carbon economy in the information age”, 2008 https://www.theclimategroup.org/sites/default/files/archive/files/Smart2020Report.pdf
3. Global e-Sustainability Initiative, The Boston Consulting Group, “GeSI SMARTer 2020: The Role of ICT in Driving a Sustainable Future”, 2012, https://gesi.org/research/gesi-smarter2020-the-role-of-ict-in-driving-a-sustainable-future
4. Global e-Sustainability Initiative and Deloitte, “Digital with Purpose: Delivering a SMARTer2030”, 2019, https://gesi.org/research/gesi-digital-with-purpose-full-report
5. S. Lange, J. Pohl, T. Santarius, “Digitalization and energy consumption. Does ICT reduce energy demand?”, Ecological Economics, Vol. 176, Oct. 2020, 106760
6. EU 5G Infrastructure Public Private Partnership Association, “5G Vision: the next generation of communication networks and services”, 2015 https://5g-ppp.eu/wp-content/uploads/2015/02/5G-Vision-Brochure-v1.pdf
7. M. Latva-aho, K. Leppänen (eds.), “Key drivers and research challenges for 6G ubiquitous wireless intelligence”, 6G Research Visions 1, Sept. 2019, ISBN 978-952-62-2353-7
8. F. Idzikowski, L. Chiaraviglio, A. Cianfrani, J. López Vizcaíno, M. Polverini, Y. Ye, “A Survey on Energy-Aware Design and Operation of Core Networks”, IEEE Communications Surveys & Tutorials, 2016, Vol. 18, No. 2, pp. 1453 – 1499
9. S. Mittal, “A survey of techniques for improving energy efficiency in embedded computing systems”, International Journal of Computer Aided Engineering and Technology, Inderscience Publishers, 2014, 6 (4), pp.440-459.
10. G. Cook, et.al., "Clicking Clean: Who Is Winning the Race to Build a Green Internet", Greenpeace study 2017.
11. M. Chiang, B.h Balasubramanian, F. Bonomi (Eds.), Fog for 5G and IoT, Information and Communication Technology Series, Wiley; 1 edition (March 27, 2017)
12. B. Kopras, F. Idzikowski, P. Kryszkiewicz, „Power Consumption and Delay in Wired Parts of Fog Computing Networks”, IEEE Sustainability through ICT Summit (StICT), 2019, 18-19 June 2019, Montréal, QC, Canada,
13. P. Kryszkiewicz, F. Idzikowski, B. Bossy, B. Kopras, H. Bogucka, "Energy Savings by Task Offloading to a Fog Considering Radio Front-End Characteristics”, IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), 8-11 Sept. 2019, Istanbul, Turkey
14. B. Bossy, H. Bogucka, “Optimization of Energy Efficiency in Computationally-Aware Adaptive OFDM Systems”, IEEE International Symposium on Personal, Indoor and Mobile Radio Communications – PIMRC’2016, 4-7 Sept. 2016, Valencia, Spain
15. B. Bossy, P. Kryszkiewicz, H. Bogucka, “Optimization of Energy Efficiency in the Downlink LTE Transmission”, IEEE International Conference on Communications, ICC 2017, 22-25 May, 2017, Paris, France
16. B. Bossy, P. Kryszkiewicz, H. Bogucka, “Energy Efficient Resource Allocation in Multiuser DF Relay Interference Networks”, IEEE Global Communications Conference: Workshops: Green and Sustainable 5G Wireless Networks, GLOBECOM 2018, 9-13 December 2018, Abu Dhabi, UAE
17. B. Bossy, P. Kryszkiewicz, H. Bogucka, „Energy Efficient Wireless Relay Networks with Computational Awareness”, IEEE Transactions on Communications, Vol. 68, No. 2, Feb. 2020, pp. 825 - 840