congestion2

The Internet and Congestion

Introduction

In the previous section of this essay the components of demand for internet usage were examined. One of the major components of demand for internet usage was the network effect. Although increased network size initially produced increasing marginal benefit, increased size can eventually lead to negative effects which will produce a declining marginal benefit of network size. If the network grows too large, the negative effects of increased size may outweigh the positive. As a result the marginal benefit of increased size will change from being a positive value to a negative value. In this section of the essay the problem of congestion will be explored more thoroughly. Congestion of internet data/traffic will be examined and it will be shown that a negative externality exists once the quantity of traffic demanded exceeds the capacity of the infrastructure carrying the data/traffic..

Points of Congestion

The flow of data/traffic on the internet can suffer from congestion at a number of points. Consider the retrieval of a web page from the US White House internet site (http://www.whitehouse.gov) by a user at Newcastle University, Australia. Possible points of congestion include the White House server's internet connection, the Australia-US link that the data travels over, the regional network which delivers the traffic to the University, the University's own connection to the internet and finally the LAN which the user is located on. Any or all of these links can suffer from congestion as a result of the traffic being carried exceeding the capacity of the connection.

Congestion of the flow of data/traffic is being worsened by the increasing use of multimedia technology, such as videoconferencing, by users who do not fully appreciate its impact on the network. This is recognised as one of the major contributors to internet traffic congestion. (Bohn et. al (1993:2-3)). In particular, "as network access becomes ubiquitous there will be an ever-increasing number of unsophisticated users who have access to applications that can cause severe congestion if not used properly." (MacKie-Mason & Varian (1997:43))

It should be noted that from the perspective of internet consumers, hardware congestion and data/traffic congestion are often indistinguishable. An internet user attempting to retrieve a file from a file repository in another country will generally be unable to tell whether the dominant cause of congestion is the hardware at the file repository, or the various network links between the repository and the user. Thus, although the distinction between hardware and data/traffic congestion is important to internet providers, the effect on internet users is generally the same . That is, they both slow down the rate of transfer of data/traffic but do not generally result in the user being excluded from the internet. For the purpose of our analysis, the congestion of hardware will be assumed to simply be an element of the congestion of data/traffic and our primary focus will remain on the congestion of data/traffic.

The Congestion Externality

As mentioned earlier, increased network size may eventually produce negative effects which reduce the benefits of increased network size and in some cases the marginal benefit of increased size can become negative. This essay will therefore now examine the existence and impact of congestion of internet data/traffic.¹

The congestion externality arises as "during congested times, one user imposes a cost in the form of a degradation of service or delays on other users when the first uses the network" (Villasis (1996)). An analogy used to explain internet congestion is that of traffic jams and delays on highways. (Hallgren & McAdams (1997:472-3)) When choosing the highway, road users only consider their marginal private cost, not the marginal social cost of their actions. At rush hour, the addition of each extra car imposes a cost, in terms of extra delay, on all other road users. This means the marginal social cost due to delays and congestion, exceeds the marginal private cost to the individual road user giving rise to a negative congestion externality.

A similar situation can be argued to exist in relation to internet data/traffic. Consider Figure 3 below.

Figure 3 illustrates the congestion externality for internet data/traffic. In Figure 3 there is an internet connection with a fixed capacity of Q# megabytes per hour. For all quantities of traffic up to Q# , the marginal cost of additional traffic is constant at P0 and no delays are experienced. Indeed, it has been suggested that on an uncongested network the marginal cost of additional traffic is close to zero. (MacKie-Mason & Varian (1994d:20,30); Hazlett (1994:3)) However once the quantity of traffic demanded exceeds Q# , all traffic is delayed.² Thus, as a user generates additional traffic past Q# they experience delays. The Marginal Private Cost MPC increases since "time spent by users waiting for a file transfer is a social cost, and should be recognised as such in any economic accounting" (MacKie-Mason & Varian (1994a:11)). Furthermore, since the additional traffic generated creates delays for all users, not just the user generating the marginal traffic, the Marginal Social Cost MSC lies above the MPC. In Figure 3, this is represented by both the MPC and MSC rising once the capacity of the link Q# is exceeded.

The negative congestion externality arises because of this divergence between MPC and MSC. Consider the demand curve D0 in Figure 3. Let us assume that the consumer pays a price equal to the marginal private cost of their additional traffic.³ With demand curve D0 the quantity demanded is Q0 and the consumer pays a price of P0. Now, let us assume that there is a positive relationship between network size and network traffic⁴. As the network size grows the quantity of data/traffic demanded will increase and eventually exceed the capacity of the network.⁵ In Figure 3 this is represented by an increase in demand from D0 to D1.

With demand curve D1 consumers now demand quantity Q1 megabytes per hour, where the price P2 equals the MPC of that quantity of traffic. However at Q1 the traffic of all users has been slowed and the MSC is equal to P1. The socially optimal level of traffic occurs where price = marginal social cost. In Figure 3 this is at price P3 and quantity Q2. Therefore at quantity Q1 a negative congestion externality of P1-P2 exists.

It is important to remember however that the increase in demand from D0 to D1 could also be bought about by other factors besides the network effect. For example, the increase in demand could be the result of changes in the price of substitutes and compliments, changes in income or changes in consumer preferences. These factors can produce variations in demand even though the network size may be constant.

Conclusion

In this section of the essay, the existence of the congestion externality has been demonstrated. It has been shown that once the capacity of the network is exceeded, all users experience delays, not just the user creating the additional traffic. A divergence between MPC and MSC thus occurs, resulting in a negative congestion externality. In the next section of this essay the determinants of the size of the congestion externality and the relationship between the congestion externality and internet demand will be explored.

Endnotes

1. As highlighted previously, anti-social behaviour and 'spam' can also reduce the marginal benefit of increased network size. However congestion is probably the area which economic analysis and solutions are most obviously applicable to.

2. The reason all traffic is delayed is due to the operation of the 'Slow Start/Congestion Avoidance/Fast Retransmit' methods implemented by TCP designed to addresses congestion of internet traffic. It aims to slow down the rate of flow of each user's traffic so that the collective sending rate is equal to the capacity at the congested point. A simplified explanation is as follows. Data is sent out initially at the rate of one 'packet' at a time. Packets are sent into the network until an episode of congestion is detected. The rate of transmission is then slowed down until the congestion is no longer detected, then gradually increased again until another episode of congestion is detected. This method ensures that network capacity is fully utililised. For a more detailed technical explanation see (Stevens (1997))

3. It should be noted that many IAP's do not use usage based pricing and where they do it might be expected that they will charge the average total cost of data/traffic so as to cover their fixed costs.

4. This assumption should be valid whether one uses number of users or number of websites as the measure of network size. If number of users is used it is reasonable to expect that an increase in users will generally lead to increased demand for usage. This will result in an increase in actual traffic unless the supply is perfectly inelastic. If web sites is the measure, increased web sites raises the marginal private benefit of usage leading to increased demand for usage and again this is likely to lead to an increase in actual traffic so long as supply is not perfectly inelastic.

5. This is of course reliant on the earlier assumption that increased network size does not result in an increase in network capacity. For example consider a typical internet access provider with a fixed size connection to the internet. As the number of web sites on the internet grows the demand for internet usage, and the amount of internet traffic, also grows. Assuming that the internet access provider does not alter the size of their connection, the increased network size will eventually mean that users will be demanding a greater amount of traffic than the capacity of the internet provider - resulting in congestion.

Last update 12-Jun-98. To visit my personal web site click here.