The energy and greenhouse-gas implications of internet video streaming in the United States

Figure 1 depicts the energy consuming processes and devices required in providing DVD video service to the in-home viewer. The process steps in figure 1 include DVDs being transported by freight from the manufacturing plant to a storefront or distribution warehouse. For rentals, DVDs are then removed from their manufacture case and placed in a company-specific case (Store Rental) or rental sleeve (Mail Rented). DVDs at distribution warehouses are then transported to the consumer through postal delivery, while DVDs at storefronts are picked up directly by the consumer. A playback device (e.g., a DVD player) is then coupled with a viewing device (e.g., a television) to watch the video.

Zoom In Zoom Out Reset image size

Table 1 presents estimates of 2011 US DVD activity that are used in this study to estimate national energy use, separated by DVD category. An approximate 2.2 million DVDs mailed per day by Netflix [13, 16] is assumed to represent the whole mail rental industry, since Netflix dominates the market as a rent-by-mail provider [17]. Each DVD is assumed to contain two hours of video content and be viewed once per mailing, resulting in an estimated 1.6 billion viewing hours annually for US mail rentals. US store rented DVDs are estimated to account for 3.6 billion viewing hours, based on the market distribution of movie rentals shown in figure 2 [17].

Zoom In Zoom Out Reset image size

US rental DVDs are assumed to have an average lifetime of two to five years, with a base-case estimate of three years, which matches the short-life property classification assigned to DVDs [18, 19]. The lifetime for purchased DVDs (i.e., before obsolescence) is extended to three to eight years, with a base-case estimate of five years, to account for reduced use and handling relative to rental DVDs.

Annual US consumer purchases of DVDs are estimated at about 1.2 billion, based on market analysis data [20]. Lacking additional data, consumer DVD sales are equally divided between mail purchased and store purchased DVDs (600 million each). For rental DVDs, one-third of the rental DVD stock is estimated to be replenished each year, assuming a near steady-state in rental DVD stock and a three year average rental DVD lifetime. Assuming each DVD is rented 20 times per year [21], the number of DVDs bought by rental companies annually in the US for the Mail Rented and Store Rented categories, is estimated at 14 million and 30 million, respectively.

Consumer purchased DVDs are assumed to be viewed an average of once per year over a five-year average lifetime, corresponding to 3.0 billion annual views from each the Mail Purchased and Store Purchased DVD stock. The sum of annual DVD viewing hours from all four categories equals approximately 17.2 billion, or a per capita viewing of about five hours per month, which is consistent with recent US consumer media usage estimates [22].

Table 2 presents DVD characteristics, embodied energy estimates, and shipping distances that are utilized as base-case values this study [13, 23, 24]. A distance of 1825 km for shipping a DVD from manufacturing plant to store location is assumed based on previous work [22] and applied to all four DVD types, assuming a broad placement of stores and distribution centers across the US. Primary energy and CO₂(e) emissions associated with transport from the manufacturing plant are based on fuel intensity estimates normalized by shipment mass for diesel fueled combination trucks [25].

A postal service transport of 210 km (round trip) is assumed to deliver the mail purchased DVDs from distribution warehouse to consumer [23] in diesel trucks, with the primary energy and CO₂(e) emissions of transport again estimated from [25]. A 17 km (round trip) distance is assumed by personal vehicle for Store Purchased, based on estimates by [24] for the average round trip distance to retail CD stores in the US. Also consistent with [24], 50% of the trip is apportioned to the DVD transport to account for multiple purchases and errands per trip in the base-case, while 90% and 10% are used in the high and low cases, respectively. Outlets for DVD rentals are assumed to be in closer proximity than DVD purchases and a round trip distance of 2.7 km is used, corresponding to the US national average household distance to the nearest supermarket [26], which is a common location for DVD rental kiosks. Energy use and emissions due to consumer travel are based on gasoline intensity estimates [25]. Vehicle fuel efficiency for a mid-size sedan, minivan, and compact hybrid vehicle are used in the base-case, high case, and low case, respectively [27]. For rental DVDs, one round trip for DVD pickup and one round trip to return the DVD is assumed, corresponding to distances of 420 km by postal service for Mail Rented DVDs and 5.4 km by consumer for Store Rented DVDs. Again, 50% of the consumer trip for Store Rented DVDs is apportioned to the DVD transport to account for multiple rentals (or renting and returning DVDs simultaneously) and additional errands conducted per trip.

Figure 3 depicts the energy using processes and devices associated with providing streaming video service to the in-home viewer. The process steps in figure 3 show video files originating from servers in data centers. The video files are then sent across the network transmission infrastructure; from a core/metro network to an access network connection pathway. The transmission then reaches a router or cable box (i.e., customer premise equipment) within the home before being sent to a playback device (e.g., set-top box) that is coupled with a viewing device (e.g., television) to watch the video.

Zoom In Zoom Out Reset image size

Table 3 presents the server and network characteristics assumed as base-case estimates for the 2011 US streaming video delivery system. Total current US streaming viewing hours for full-length movies or television programs are estimated at 3.2 billion h, which is based on the Rental DVD viewing hour estimates in table 1 and the distribution of movie rental shares shown in figure 2 [17], and is consistent with other market analysis data [20]. An average streaming rate of 2.33 Mbps [28] is applied to the annual hours of video content to estimate the total bits of streamed content. To allocate server power (including idle power) across data flows, each server is assumed to draw an average power of 300 W and stream content at a maximum sustained rate of 1.5 Gbps based on the configuration of Netflix content delivery network devices [29]. Data centers hosting streaming video are assumed to have the characteristics of cloud data centers, where the number of servers in use scales with demand [30], allowing the number of servers to be estimated by applying the bits processed per server to the total bits streamed annually of video content. A power use effectiveness of 1.3 is assumed to represent the efficient use of infrastructure equipment in larger centers [5, 31]. When including an average utilization rate of 40% for servers in cloud data centers [32], approximately 1400 dedicated servers are estimated to meet current US streaming video demand with an average data streaming electricity intensity of 1.1 Wh GB⁻¹. Given the large storage capacity of servers configured for video streaming [33], video storage demand is assumed to be met solely by the servers and no external hard disk drives (HDDs) within data centers are included in this analysis. Cradle-to-gate primary energy and CO₂(e) emission factors of 13.65 MJ and 0.69 kg CO₂(e) for each kWh of electricity consumed during operation are used as base-case estimates to represent a national average electricity grid mix and account for the energy and emissions during fuel extraction, refining, and transport as well as combustion and transmission losses [34]. The embodied energy and CO₂(e) intensity of data center IT equipment are derived from values in the CLEER model, which are based literature estimates [35–37] and are presented in the SOM.

This analysis uses an estimated network energy intensity range of 0.15 to 0.43 kWh per GB of data transferred across the core/metro, access, and customer premise equipment, representing a range consistent with literature values observed in [38–42]. Specifically, a base-case estimate of 0.29 kWh GB⁻¹ is derived from measurements in [38] from 2010 and 2013, and scaled to 2011 based on a 20% annual reduction in network energy intensity observed in that same study. The low case estimate of 0.15 is based on 2009 network energy intensity measurements from [42] of transmission equipment between large institutions. This low case 2009 estimate from [42] is again scaled to 2011 based on a 20% annual reduction in network energy intensity observed in [38]. The range between the base-case and the low case was used to define a high case value of equal range from the base-case, at 0.43 kWh per GB transferred. Overall embodied energy and CO₂(e) intensity across the core and access network is estimated at 0.3 MJ and 30 g CO₂(e) per GB transferred, again based on literature values utilized in the CLEER model [36, 40, 43] and presented in the SOM.

Table 4 presents the assumed time use distribution of video viewing devices for both DVD and streaming video. Personal computers are assumed to make up only a small portion of DVD viewing, estimated at 10% [44] and equally split between laptop use and using a desktop computer with a monitor. The remaining 90% of DVD viewing is assumed to occur on a television accompanied with a DVD playing device. Along with DVD players, video game consoles are also assumed to operate as playback devices for DVDs [45]. The 90% of DVD playback time from these devices is apportioned at 14% and 76% for game consoles and DVD players, respectively, based on annual hourly electrical appliance usage data [46] and estimates of video games console use dedicated for playing DVDs [47].

The percentages of total video streaming viewing time attributable to computers, televisions, and mobile devices are estimated at 20%, 77%, and 3%, respectively [48]. Similar to the DVD viewing assumptions, computer use is split equally between using a laptop and using a desktop computer with a monitor. Internet enabled television viewing only accounts for 6% of the playback device time for video streaming [49], while the remaining television viewing time is assumed to be paired with another device for video playback. Video playback through game consoles with internet access is assumed to account for 50% of all streaming video playback hours [45], with the remaining 21% of video playback hours applied set-top boxes, which are assumed to include internet-enabled DVD players [49].

Operational energy use for each of the end-user devices in table 4 is based on US DOE Building Energy Data Book values [46] in the base-case estimate. The base-case values are increased or decreased by 50% to represent the high and low case estimates, respectively, which is consistent with the efficiency ranges observed comparing Building Energy Data Book values with other estimates [50] and US EPA Energy Star products criteria [51]. Operational energy use is distinguished for devices in 'on' mode, 'off' mode, and 'idle' mode, when applicable. Idle- and off-mode energy use is allocated to the viewing time based on the proportion of device on-time applied to video viewing. The embodied energy and CO₂(e) intensity of end-user devices is again based on literature estimates contained in the CLEER model [37, 52–54] and presented in the SOM.

The primary energy use and CO₂(e) emissions estimates presented in the base-case assume electricity generation from US average electricity grid mix. While the US electricity grid mix is appropriate for estimating the average primary energy use and emissions per viewing hour for all US residents in the base case, how each viewing method compares at the local level for any US individual is highly dependent upon the local electricity grid mix. For example, a household located in the state of Washington is supplied with electricity generated largely by hydropower, whereas a household located in the state of Wyoming relies on electricity generated by coal-fired power plants. The primary energy and CO₂(e) intensities of these two electricity mixes differ greatly. The effect of local grid mix on US individual results has been assessed in figure S1 in the SOM, which presents results for the primary energy use and CO₂(e) emissions per viewing hour for these two drastically different electricity mixes. The results demonstrate how differences in electricity mix lead to differences in the primary energy use and emissions associated with a viewing hour, as well as relative differences in the contributing factors to primary energy use and emissions. In areas like Washington with significant hydropower, CO₂(e) emissions are reduced across all viewing methods, but the comparison leans decidedly away from video viewing associated with consumer driving, where Store Rented viewing produces four times more CO₂(e) emissions per viewing hour than video streaming. In fossil-fuel dominated electricity areas like Wyoming, the higher emissions electricity increases emissions across all viewing methods but the relative differences among the viewing methods begin to diminish, with Store Rented viewing producing about 50% more CO₂(e) emissions per viewing hour than video streaming. See figure S1 and the accompanying text in the SOM for additional detail regarding this comparison. While this paper applies an average US electricity grid mix to examine the net effects of shifting from DVDs to streaming video at the national level, future work should consider how variations in local electricity grid mix impact the comparison of video viewing methods for any individual. Future research should also explore how electricity generation trends toward lower greenhouse-gas intensive fuels and renewables could help to reduce the national-level impacts of streaming video in the future.

While video streaming quality is rapidly improving, DVDs still offer a higher quality video and audio service (in the form of BluRay discs). While Netflix's maximum Super HD streaming rate is 6 Mbps, the typical streaming rate of a BluRay disc is approximately 20 Mbps. The analysis presented here does not account for this difference in utility and instead considers all video viewing methods as equivalent services to the consumer. The quality gap between streaming and DVD video may continue to close with improved data compression technology, however, if more complex video platforms gain popularity in the marketplace (i.e., ultra high-definition or three-dimensional video), streaming video may require greater data transfer rates while the change in manufacturing and transport of physical media would be minimal. Determining how improved levels of quality affect streaming and DVD video differently, and developing methods to normalize those differences, are opportunities for future research.

Network energy use in this analysis is based on a traditional core/metro/access network architecture and does not consider video distribution that retrieves content from various strategically located third-party content delivery networks. Additionally, growth in mobile device use and improvements in cellular data transmission may lead to more video viewing across cellular networks. Future analyses should explore how shifts in data transmission networks for video content affect overall energy use of video streaming.

The operational energy demands of end-user devices account for the majority of primary energy use and CO₂(e) emissions associated with both DVD and streaming video viewing. The energy demand for streaming end-user devices is much less than DVD viewing, owing to a shift away from less efficient DVD players. However, the potential for increased efficiency gains from streaming end-user device is still significant. Set-top boxes and video game systems, which support the majority of video streaming content, still include older less efficient models that require the same amount of power whether in active or idle modes [46]. Internet-enabled televisions, as well as new and emerging set-top boxes (e.g., Apple TV and Roku) that use drastically less power [59, 60], portend the possibility of significant improvements in video streaming efficiency. Future analyses should account for these newer end-user video streaming devices as they become more prevalent.

While the scope of this analysis is limited to comparing video streaming with DVD videos, the growth of video streaming can also be expected to serve as an alternative to other forms of video entertainment, such as cable and broadcast network television. Energy and CO₂(e) analyses of these more traditional forms of television are needed to better understand the implications of video streaming beyond DVD replacement. Additionally the convenience, selection, and quality of video streaming may also lead to more video viewing hours. Future research should account for possible absolute increases in total viewing and compare this increase in video-related energy demand with the energy intensities of activities that video viewing may be replacing.

The network electricity intensity used in this analysis is derived by normalizing the total electricity of the network system by the amount of data transferred over a period of time. While the kWh GB⁻¹ ratio implies linearity, in reality the network electricity remains primarily static during incremental increases in data transmission. Significant increases in electricity demand are only observed when greater data traffic requires an expansion of the network with additional equipment. Consequently, a given network with greater data transmission results in a lower network electricity intensity. Similarly, since equipment energy use throughout our analysis (from data center servers to end-user devices) includes the idle energy associated with that equipment, increasing the viewing hours on an end-user device reduces the number of idle hours apportioned to each viewing hour. This method causes the per-hour viewing energy to reduce as more videos are viewed per device. Developing other methods to account for network intensity and idle energy use, such as apportioning access networks and end-user device energy by household, may yield different insights and provides an opportunity for future research.

The LCA applied in this research is limited to energy and CO₂(e) emissions. Shifting from DVDs to video streaming may also lead to other environmental burdens due to changes in life-cycle resource use and emissions. Future LCAs of video streaming should include other environmental and human health impact categories for a broader assessment of environmental burdens and shifts between them.