It has long been recognized that networked services contain weak-link vulnerabilities. That is, the performance of any frontier device depends on the performance of every contributing component and service. This column focuses on one such phenomenon, which goes by the label "congestion." No, this is not a new type of allergy, but, as with a bacteria, many users want to avoid it, especially advanced users of frontier network services.
Congestion arises when network capacity does not provide adequate service during heavy use. Congestion slows down data delivery and erodes application performance, especially for time-sensitive apps such as movies, online videos, and interactive gaming.
Concerns about congestion are pervasive. Embarrassing reports about broadband networks with slow speeds highlight the role of congestion. Regulatory disputes about data caps and pricing tiers question whether these programs limit the use of data in a useful way. Investment analysts focus on the frequency of congestion as a measure of a broadband network's quality.
What economic factors produce congestion? Let's examine the root economic causes.
Congestion arises when demand for data exceeds supply in a very specific sense.
Start with demand. To make this digestible, let's confine our attention to US households in an urban or suburban area, which produces the majority of data traffic.
No simple generalization can characterize all users and uses. The typical household today uses data for a wide variety of purposes—email, video, passive browsing, music videos, streaming of movies, and e-commerce. Networks also interact with a wide variety of end devices—PCs, tablets, smartphones on local Wi-Fi, streaming to television, home video alarm systems, remote temperature control systems, and plenty more.
It is complicated, but two facts should be foremost in this discussion. First, a high fraction of traffic is video—anywhere from 60 to 80 percent, depending on the estimate. Second, demand peaks at night. Most users want to do more things after dinner, far more than any other time during the day.
Every network operator knows that demand for data will peak (predictably) between approximately 7 p.m. and 11 p.m. Yes, it is predictable. Every day of the week looks like every other, albeit with steady growth over time and with some occasional fluctuations for holidays and weather. The weekends don't look any different, by the way, except that the daytime has a bit more demand than during the week.
The bottom line: evenings require far greater capacity than other times of the day. If capacity is not adequate, it can manifest as a bottleneck at many different points in a network—in its backbone, in its interconnection points, or in its last mile nodes.
This is where engineering and economics can become tricky to explain (and to manage). Consider this metaphor (with apologies to network engineers): Metaphorically speaking, network congestion can resemble a bathtub backed up with water. The water might fail to drain because something is interfering with the mouth of the drain or there is a clog far down the pipes. So, too, congestion in a data network can arise from inadequate capacity close to the household or inadequate capacity somewhere in the infrastructure supporting delivery of data.
Numerous features inside a network can be responsible for congestion, and that shapes which set of households experience congestion most severely. Accordingly, numerous different investments can alleviate the congestion in specific places. A network could require a "splitting of nodes" or a "larger pipe" to support a content delivery network (CDN) or could require "more ports at the point of interconnection" between a particular backbone provider and the network.
As it turns out, despite that complexity, we live in an era in which bottlenecks arise most often in the last mile, which ISPs build and operate. That simplifies the economics: Once an ISP builds and optimizes a network to meet maximum local demand at peak hours, then that same capacity will be able to meet lower demand the rest of the day. Similarly, high capacity can also address lower levels of peak demand on any other day.
Think of the economics this way. An awesome network, with extraordinary capacity optimized to its users, will alleviate congestion at most households on virtually every day of the week, except the most extraordinary. Accordingly, as the network becomes less than awesome with less capacity, it will generate a number of (predictable) days of peak demand with severe congestion throughout the entire peak time period at more households. The logic carries through: the less awesome the network, the greater the number of households who experience those moments of severe congestion, and the greater the frequency.
That provides a way to translate many network engineering benchmarks—such as the percentage of packet loss. More packet loss correlates with more congestion, and that corresponds with a larger number of moments when some household experiences poor service.
Tradeoffs and Externalities
Not all market participants react to congestion in the same way. Let's first focus on the gazillion Web firms that supply the content. They watch this situation with a wary eye, and it's no wonder. Many third-party services, such as those streaming video, deliver a higher-quality experience to users whose network suffers less congestion.
Many content providers invest to alleviate congestion. Some invest in compression software and superior webpage design, which loads in ways that speeds up the user experience. Some buy CDN services to speed delivery of their data. Some of the largest content firms, such as YouTube, Google, Netflix, and Facebook, build their own CDN services to improve delivery.
Next, focus on ISPs. They react with various investment and pricing strategies. At one extreme, some ISPs have chosen to save money by investing conservatively, and they suffer the complaints of users. At the other extreme, some ISPs build a premium network, then charge premium prices for the best services.
There are two good reasons for that variety. First, ISPs differ in their rates of capital investment. Partly this is due to investment costs, which vary greatly with density, topography, and local government relations. Rates of investment tend to be inherited from long histories, sometimes as a product of decisions made many years ago, which accumulated over time. These commitments can change, but generally don't, because investors watch capital commitments and react strongly to any departure from history.
The second reason is more subtle. ISPs take different approaches to raising revenue per household, and this results in (effectively) different relationships with banks and stockholders, and, de facto, different budgets for investment. Where does the difference in revenue come from? For one, competitive conditions and market power differ across neighborhoods. In addition, ISPs use different pricing strategies, taking substantially different approaches to discounts, tiered pricing structures, data cap policies, bundled contract offerings, and nuisance fees.
The use of tiers tends to grab attention in public discussion. ISPs segment their users. Higher tiers bring more bandwidth to a household. All else equal, households with higher tiers experience less congestion at peak moments.
Investors like tiers because they don't obligate ISPs to offer unlimited service and, in the long run, raise revenue without additional costs. Users have a more mixed reaction. Light users like the lower prices of lower tiers, and appreciate saving money for doing little other than email and static browsing. In contrast, heavy users perceive that they pay extra to receive the bandwidth that the ISP used to supply as a default.
ISPs cannot win for losing. The archetypical conservative ISP invests adequately to relieve congestion some of the time, but not all of the time. Its management then must face the occasional phone calls of its users, which they stymie with phone trees that make service calls last 45 minutes. Even if users like the low prices, they find the service and reliability quite irritating.
The archetypical aggressive ISP, in contrast, achieves a high-quality network, which relieves severe congestion much of the time. Yet, such firms (typically) find clever ways to pile on fees, and know how to stymie user complaints with a different type of phone tree that makes calls last 45 minutes. Even when users like the quality, the aggressive pricing practices tend to be quite irritating.
One last note: It is a complicated situation where ISPs interconnect with content providers. Multiple parties must invest, and the situations involve many supplier interests and strategic contingencies.
Some observers have alleged that the biggest ISPs have created congestion issues at interconnection points for purposes of gaining negotiating leverage. These are serious charges, and a certain amount of skepticism is warranted for any broad charge that lacks specifics.
Somebody ought to do a sober and detailed investigation to confront those theories with evidence. (I am just saying.)
What does basic economics tell us about congestion? Congestion is inevitable in a network with interlocking interests. When one part of the network has congestion, the rest of it catches a cold.
More to the point, growth in demand for data should continue to stress network capacity into the foreseeable future. Since not all ISPs will invest aggressively in the presence of congestion, some amount of congestion is inevitable. So, too, is a certain amount of irritation.