- The Cost of Network Latency - Network latency in the U.S. has risen steadily with the deployment of more communications networks and increased complexity of internetworking. Just as air travel involves more segments and connections, internet transport involves more hops and cross-connects with similar consequences for end-users: evergrowing delays. We estimate that latency in the U.S. will grow from roughly 4 ms in 213 to nearly 7 ms by 217 on average across fixed and mobile networks 123. 14 Latency Trends, U.S. 12 1 Latency (ms) 8 6 4 42 47 53 59 67 2 213 214 215 216 217 Fixed Mobile Blended The effects of latency we believe are profound, and its underlying causes are attributable largely to the unstructured way in which the Internet has evolved physically and logically. In the following sections, we describe the fundamental drivers of latency, its potential economic consequences for the internet, and possible solutions for network and content providers. Technology Foresight Latency can be understood in the context of fiber transport networks, which comprise federations of smaller sub-networks. These federations are maintained by peering arrangements between service providers, such as AT&T and Verizon, and often are enabled by third party data center providers or meet-me points. 1 FCC, Measuring Broadband in America, 213 and 211: Mean wireline latency grew by 23% between 211 and 212, from 3 ms to 37 ms. We assume conservatively that latency grows at 1% annually in our base-case scenario, 5% in our best-case scenario, and 2% in our worst-case scenario. 2 OpenSignal: Mobile latency ranges from 8 ms to 1 ms for LTE networks, and from 1 ms and 2 ms for 3G networks. We adopted the low end of the LTE range for our analysis. 3 Blended latency is a weighted average of fixed and mobile latency based on fixed and mobile e- commerce sessions. incode Consulting Page 1
Fiber networks can be divided into access, metro, and long-haul segments, each with unique conduit, pole-top, and right of way restrictions. Segments often are managed by and interconnected to different operators. For example, long distance might be provisioned by AT&T, metro connectivity by Zayo, and access by Cox, with interconnection provided by an independent meet-me operator, such as InterNap. Compounding physical network complexity are suboptimal meet-me point locations, as well as complex rights-of-way rules enforced by federal, state, and local agencies. Municipalities, in particular, can impose strict regulations that effect conduit routes and infrastructure, often creating transport inefficiencies for local traffic. For example, conduits often follow high internet business traffic corridors, and interconnect with residential fiber networks at meet-me points collocated at cable head-ends. As a result, there is a mish-mash of interconnected fiber and peering arrangements that increase the distance traveled on the internet to hundreds or even thousands of miles, adding 1 ms for every 1 miles of transport. The primary drivers of latency include cross-connects, fiber transport circuits, and traffic processing equipment, such as switches and routers. Cross-connects introduce two forms of latency: Distance-related and interconnection-related. Distance-related delays are caused by the distance traveled to the meet-me point. Interconnection-related latency is caused by conversions of packet data from optical to electrical and back when transiting colocation facilities. Latency at a meet-me point or data center further depends on the amount of access, metro, and core routing required, potentially adding precious milliseconds of delay to the transport of information. Fiber latency is exacerbated by the physical location of last mile gateways, fiber taps, manholes, and other cross connection points. Conduit-imposed route miles can add further delay and routing complexity for both IP signaling and bearer traffic. Finally, equipment imposes processing delays through optical multiplexors, electrical multiplexors, routers, servers, and other hardware components. This gives rise to new ways of designing intra-data center and rack-to-rack data flow, using new technologies such as Software Defined Networking and Open Flow architecture. AT&T and others recently announced plans to deploy such technologies to further enhance their network performance. Business Insight To estimate the cost of latency, we developed a simplified model capturing two key drivers of network delay: Transport and Interconnect: incode Consulting Page 2
We then estimated the consequences of latency for e-commerce spending in the U.S., which in 213 exceeded $38 B 4. e-commerce Revenue, U.S. 7 ecom Revenue, $B 6 5 4 3 2 386 429 477 531 591 1 213 214 215 216 217 Fixed, U.S. (cgr: 1.6%) Mobile, U.S. (cgr: 35.%) To understand how latency impacts e-commerce, we modelled its effect on adoption, usage, and abandonment rates among fixed and mobile e-commerce users. 4 emarketer, Forrester incode Consulting Page 3
We considered best- and worst-case scenarios for latency, and correlated these scenarios to e-commerce spending by applying the following assumptions: Abandonment grows by 1% for every 1 ms of latency 5 Usage and adoption each decline by 2% for every 1 ms of latency 6 The following chart shows the resulting correlation between e-commerce spending and latency, based on 217 projections: 5 Amazon estimates that revenue declines by 1% for every 1 ms of latency. We assume this reflects the latency-induced abandonment rate experienced at the Amazon.com website. 6 Google estimates that traffic declines by 4% for every 1 ms of latency. We attributed 5% of this decline to e-commerce usage, and 5% to e-commerce adoption. incode Consulting Page 4
Latency Impact on e-commerce Revenue (217 est.) Fixed e-commerce Revenue, 218 ($B) 59 23.8 585 23.6 23.4 58 23.2 575 23. 57 22.8 565 22.6 22.4 56 22.2 555 22. 5 1 15 2 Latency (ms) Fixed, 217 ($1B per 3 ms) Mobile, 217 ($1B per 122 ms) Mobile e-commerce Revenue, 218 ($B) As this chart shows, fixed e-commerce spending declines by $1B for every 3 ms of latency, while mobile e-commerce spending declines by $1B for every 122 ms of latency. As a result, we estimate that lost e-commerce spending will grow from $8B in 213 to nearly $2B in 217 (2% and 3% of total e-commerce spending, respectively), or between $5B and $8B cumulatively over five years, as illustrated in the following chart. Annual Lost ecommerce Revenue ($B) 2 18 16 14 12 1 8 6 4 2 Lost e-commerce Spend 18 9 8 15 7 12 6 1 5 8 4 3 2 1 213 214 215 216 217 Cumulative, Range Fixed, Base Case Mobile, Base Case Cumulative, Base Case Cumulative Lost ecommerce Revenue ($B) Two additional insights emerged from our work. First, losses accelerate during this period as the mix of higher-latency mobile traffic grows alongside increased mobile adoption. Second, spending may be inhibited more by the reluctance among users to engage in e-commerce due to poor network performance, than incode Consulting Page 5
by the propensity of users to abandon during active e-commerce sessions; that is, negative e-commerce experiences may have the larger and longer-lasting consequence of discouraging usage and adoption. Our analysis excludes other important latency impacts. In the financial sector, for example, small network delays can generate large trading losses; in cloud computing, latency can discourage adoption and usage among business users. It is difficult to estimate the full impact of latency, but based on this high-level analysis, we believe actual costs could exceed our base-case estimates by a factor of two or more in the U.S. Conclusions Latency poses a growing dilemma for network operators and content providers. We believe this problem can be mitigated by enhancing the underlying technology infrastructure, and defining new business models that monetize lowlatency internetworking. Enhancing the infrastructure will involve fundamentally rethinking the underlying physical and logical architecture of internetworks, including cross connects, conduit design, and peering. This may give rise to new plant, software, topology, and data center strategies. Several industries, notably financial services, have taken proactive measures to build custom low-latency networks 7. Furthermore, in 4G wireless networks, critical inter base-station signaling requirements for small cell network deployment require low latency solutions in order to successfully implement new standards, such as inter-cell interference coordination and self-organizing networks 8. It is clear that future fiber networks will require increasingly intelligent design strategies, and that the key network corridors and existing networks will need to be reviewed, enhanced, and possibly decommissioned and replaced to mitigate latency issues. Latency also creates the potential for new business models, particularly as net neutrality restrictions are relaxed. For example, network operators can charge for premium, non-stop routes that circumvent cross-connects and follow lesscongested corridors to minimize delays and enhance customer experience. Such models already are gaining momentum, as evidenced by Comcast s recent agreement to deliver Netflix s video traffic more expediently to end-users. We believe such arrangements can benefit all parties, as users gain an improved experience, content providers benefit from greater adoption and usage, and network operators generate the economic returns needed to fund further network expansion. 7 Wall Street Journal, High-Speed Stock Traders Turn to Laser Beams, February 214 8 Metro Ethernet Forum incode Consulting Page 6
Latency poses both a challenge and an opportunity for the broader internet ecosystem that can be solved through a mix of technology and business model innovation. Through careful planning and implementation, we believe the industry and end-users alike will derive greater value from the ever-expanding internet. About Peter Atwal and David Waite are Managing Directors at incode Consulting, a strategy and technology consulting firm focused on the communications, media, and technology segments. We help companies and investors to understand industry trends, develop strategies and business models to accelerate growth, and define innovative ways to maximize profitability and shareholder returns. Learn more at www.incodeconsulting.com. incode Consulting Page 7