Investigating the Inconsistencies among Energy and Energy Intensity Estimates of the Internet: Metrics and Harmonizing Values
The report performs an in-depth analysis of estimates of the energy intensity (EI) and the overall energy (E) consumption of the Internet. The two values should be connected via the overall Internet traffic (T), as multiplying EI by T should yield E. This identity, however, is not satisfied by the existing assessments today; on the contrary, the results obtained via the two methods lie a factor of 5-26 apart. The study identifies two important sources for this inconsistency: the different treatment of access networks (ANs), and the inherent biases towards overestimates and underestimates that seem to reside within top-down and bottom-up modelling, respectively. Considering the wide-area network (WAN), fixed access networks (FANs), and radio access networks (RANs) individually, the study shows that, rather surprisingly, the largest discrepancies do not stem from the fairly recent and quite heterogeneous RANs, but from the well-established WAN.
The study further identifies three levels of harmonisation between EI and E; in order of rigour and precision, but also of increasing challenges and uncertainties, they are: i) a global one across all types of networks, ii) one that individually harmonises within the WAN and within the access networks, and iii) one harmonising within each of the three types of the networks. Given the incompatibility of energy and energy intensity estimates for the WAN within existing literature, only a global harmonisation could be achieved. Following from this mutual validation, which also tried to harmonise as much as possible within the individual types of networks, a fair approximation of today’s energy and energy intensity values is as follows: the energy consumption of the WAN in 2020, E2020 (WAN) = 110 TWh, E2020 (FAN) = 130 TWh, and E2020 (RAN) = 100 TWh; the energy intensity of the WAN in 2020, EI2020 (WAN) = 0.02 kWh/GB, EI2020 (FAN) = 0.07 kWh/GB, and EI2020 (RAN) = 0.2 kWh/GB. For short-term extrapolations to the future, yearly energy intensity reduction factors of 0.8 for both WAN and RAN, and 0.85 for the FAN, appear meaningful and rather conservative.
Future efforts are needed in particular to harmonise energy assessments of the WAN. More consistent results can be achieved by either deploying both top-down and bottom-up methods, aiming for a hybrid method, or at least validating the results via the other method. Particularly important and often misinterpreted are the overall averages of the energy intensity of the Internet. For consistency, they need to account for i) the additive nature of the intensities of the WAN and the access networks, ii) the alternative access via FAN or RAN, and iii) the (rapidly changing) shares of these two types of access networks. The study provides equations that can be used both to compute this average, and to verify the consistency between estimates of EI and E.
Search for the Publication In: