Client-side Energy Costs of Video Streaming Oche Ejembi School of Computer Science University of St Andrews, Fife, UK Email: ooe@st-andrews.ac.uk Saleem N. Bhatti School of Computer Science University of St Andrews, Fife, UK Email: saleem@st-andrews.ac.uk Abstract—Through measurements on our testbed, we show how users of Netflix could make energy savings of up to 34% by adjusting video quality settings. We estimate the impacts of these quality settings on the energy consumption of client systems and the network. If users exercise choice in their video streaming habits, over 100 GWh of energy a year could be saved on a global scale. We discuss how providing energy usage information to users of digital video could enable them to make choices of video settings to reduce energy usage, and we estimate savings on associated electricity costs and carbon emissions. I. I NTRODUCTION Video-on-Demand (VoD) is by far the most popular type of traffic on the Internet and its use is growing. According to the Sandvine Global Internet Phenomena report for 1H 2014 [1], VoD was responsible for 64% of all the downstream traffic experienced on fixed networks, and 40% on mobile networks in the US, with similarly large proportions experienced in Europe, Asia-Pacific, and Latin America. Cisco estimates that by 2018, 79% of all the world’s Internet traffic will be some form of video [2]. However, video consumes more resources – network capac- ity, device CPU utilisation, memory, I/O, disk space etc. – than the other popular uses of the Internet, such as text, audio and still images. So, video also consumes more energy. A. Contributions and Structure We measure the client-side energy usage of the popular video-on-demand service provider Netflix to make global Fermi estimates of how much energy could be consumed by users of this particular service. We make suggestions on how this energy usage could be reduced or made more sustainable. Our contributions are: 1) Using a simple measurement methodology, we make an empirical assessment of the client-side energy usage of the world’s most popular premium VoD provider. We then make global estimates of how much energy is consumed by this service’s subscribers. 2) We examine the network characteristics of this video application and investigate the energy impacts of the traffic at the client side. 3) Finally, we discuss ways in which these users could be empowered to make choices towards saving energy while using VoD services, based on measured energy usage information, and feedback to the users. In Section II, we discuss recent, related work. This is followed by a presentation of our experiment methodology, testbed and measurement strategies in Section III, with the results from our experiments in Section IV. We analyse the results in Section V, including a discussion of the implications for real-world scenarios in Section VI. Finally, we conclude and present avenues for future work in Section VII. II. RELATED WORK Research in Green ICT and energy efficient systems now covers a wide range of topics. There have been high-level, analytical investigations of ICT energy usage e.g. Somavat et el. [3] for energy-efficient data- centres and servers [4], to user studies towards reducing energy usage in ICT [5]. Considering the scale of Internet traffic attributed to video, a surprisingly small amount of work has specifically looked at energy usage of multimedia, especially for non-mobile systems. Non-mobiles are responsible for the bulk of video traffic – 60% according to Nielsen [6] – and so also for the associated energy use and carbon emissions. In our previous work [7], we performed an empirical investigation into the energy usage for video decoding and encoding for different popular codecs on a commodity desktop hardware system. We observed a difference of up to a factor of 3 between software codecs for decoding, even at the same picture size (resolution) and bit rate, for our desktop client system. Similarly, Shehabi et al. [8], make an analytical life-cycle assessment of the energy used by streaming video in the US. They identify that the strongest opportunities for significant gains in energy efficiency are at the end-user level and the network. A considerable amount of research has investigated vari- ous aspects of popular multimedia applications like Netflix, YouTube and BBC iPlayer, similar (in principle) to our focus in this paper e.g [9] and [10]. However, very little work has focused explicitly on the energy usage of these services and applications, and the few that do consider such issues focus on mobile devices with the aim of extending battery life [11]. We take the position that since the bulk of video traffic is consumed by non-mobile devices (according the Sandvine report [1] and Nielsen [6]), it is important to investigate energy usage of non-mobiles, especially as those systems will not have the same level of design or engineering optimisation for energy efficient operation. As the popularity of VoD continues to increase, so will the energy cost and any associated carbon emissions. However, where mains power is easily available, there is the danger that energy consumption by VoD could