The demand on communication networks has increased over the past years and is predicted to continue for the foreseeable future [Cis16]. Cellular network access with a compound annual growth rate (CAGR) of 53 % is the main area of growth [Cis16]. This affects the network quality, bringing current network technologies to their limits [Qua13]. Future network standards like 5G promise to satisfy this demand, providing a 1000-fold increase in data rates and latencies as low as 1 ms [Qua13].

With information and communications technology (ICT) causing 10 % of the global energy consumption [Mil13], the increasing demand is also reflected in a growing energy consumption of communication networks [BBD+11]. The major contributor to the network power consumption are home gateways (HGWs) in the fixed access network, and mobile base stations in the cellular network [VHD+11]. This trend is predicted to continue [BBD+11].

To assess and optimize the power consumption of communication networks, power models of the involved devices are required. Using these, the efficiency of proposed optimization approaches can be assessed before deployment. A number of power models of conventional network equipment for different device classes can be derived from literature. Still, models of new device classes such as single-board computers (SBCs) and OpenFlow switches are not available. For each class, representative power models of several device types are presented. Further, the power consumption caused by new communication protocols such as MultiPath TCP (MPTCP) is not fully analyzed yet. This work is, to the best of the author’s knowledge, the first to publish SBC and OpenFlow power models and contributes to the understanding of MPTCP power consumption during constant bit rate (CBR) streaming.

For the analysis of the power consumption, also the knowledge of network performance is required, as it defines relative costs and the maximum number of supported users. This is well known and comparatively simple in fixed networks, but more challenging in a wireless context. A number of approaches are described in literature and implemented as commercial software (e.g. [SSM13; OpS]), but the data required for analysis and optimization is not available. Hence, extensive measurements of the cellular network are conducted in this work. The location-based availability and performance of cellular and WiFi networks are assessed in a crowd-sensing study. Based on measurements on regional trains, the predictability of the cellular service quality based only on available network technology and latency is shown to be feasible. Anomalies observed within the crowd-sensing data are analyzed using dedicated, stationary measurements. The main observation is that network management decisions have significant effects on end-to-end performance. By allocating users to random points of presence (PoPs)/exit gateways of the mobile network operator (MNO), the latency compared to the best observed allocation is increased by more than 58 % in over 80 % of the time.

Combining the energy models and network performance measurements as presented in this work, an energy evaluation environment is created to analyze the cost of mobile data communication. This combines the empirically determined performance of cellular and WiFi networks with the energy models of smartphones and traffic traces recorded by the participants of a crowd-sensing study. Thereby, the power consumption of the generated data patterns is established, and the effectiveness of network optimization approaches as presented in literature assessed. These prove to be less potent than originally claimed by the authors. This is expected considering the improvements in cellular networks and smartphones. Nonetheless, energy savings are observed. Considering the requirement of 5G networks to reduce latency to 1 ms, and improve capacity by a factor of 1000, while simultaneously reducing energy consumption, also changes in fixed access networks are required. A promising approach assuming further virtualization of networks using software defined networking (SDN) and network functions virtualization (NFV) is the placement of services closer to the end-users. Extrapolating the trend of increasing hardware capabilities of HGWs at almost constant cost, these may be used to provide additional services to local users. This may be achieved by e.g. using virtualized content distribution network (CDN) nodes running on HGWs, thus utilizing these often idle resources. This further equalizes the traffic within the core network by providing content locally and refreshing it during less traffic intensive periods. Simultaneously, the end-user perceived service quality is expected to increase. Thus, installed capacities can be used longer, resulting in better service quality at fixed energy cost.