Neutral and Non-neutral Countries in a Global Internet: What Does It Imply?

Abstract

Network neutrality is being discussed worldwide, with different countries applying different policies, some imposing it, others acting against regulation or even repealing it as recently in the USA. The goal of this paper is to model and analyze the interactions of users, content providers, and Internet service providers (ISPs) located in countries with different rules.

To do so, we build a simple two-regions game-theoretic model and focus on two scenarios of net neutrality relaxation in one region while it remains enforced in the other one. In a first scenario, from an initial situation where both regions offer the same basic quality, one region allows ISPs to offer fast lanes for a premium while still guaranteeing the basic service; in a second scenario the ISPs in both regions play a game on quality, with only one possible quality in the neutral region, and two in the non-neutral one but with a regulated quality ratio between those.

Our numerical experiments lead to very different outcomes, with the first scenario benefiting to all actors (especially the ones in the relaxed-neutrality region) and the second one mainly benefiting mostly to ISPs while Content Providers are worse off, suggesting that regulation should be carefully designed.

1 Introduction

The network neutrality debate has been raging for close to two decades and is still a very sensitive issue worldwide. Basically, network neutrality is “the principle that traffic should be treated equally, without discrimination, restriction or interference, independent of the sender, receiver, type, content, device, service or application”. This type of definition was introduced by the Federal Communications Commission, the regulator in the US in 2005, and by the European Union in 2014, among others. It is part of the principles for an open Internet according to which resources available on the Internet should be easily accessible to all entities. The debate has been highlighted in 2005 with Ed Whitacre, CEO of the Internet service provider (ISP) AT&T, complaining that distant content providers (CPs) were using his network without financially participating to its infrastructure maintenance and upgrade, while at the same time the proportion of telecommunications economy coming from advertisement and going to CPs was increasing. The threat to differentiate traffic or even block some services raised a lot of protests. Since, numerous attempts to discriminate traffic have been observed, such as ISP Madison River Communications fined in 2005 for preventing its clients from using VoIP in competition with its own “voice” offer, Comcast blocking in 2007 BitTorrent (P2P) traffic, or the recent exclusion of some traffic from data caps in wireless subscription offers (the so-called zero rating). For more on Net neutrality and its history, the reader is advised to look at [5, 7,8,9, 11, 13, 14] and references therein.

An important issue barely addressed in the literature is that while neutrality principles are imposed in many countries, it is not the case everywhere. As of March 2019, we can define two, and even three sets of countries regarding net neutrality¹:

• Countries having passed laws to protect neutrality; this includes all European Union, Canada, most of South America, Japan, India, etc. Remark though that rules are more or less strict depending on the country; for example some authorize sponsored data (that is, the possibility for content providers to pay for their traffic and exclude it from the users data cap) while other don’t.

• Countries against neutrality as recently the USA, or other big countries such as Russia, China, etc. Claimed reasons are not always the same: economic efficiency in the USA, control of content/traffic by deep packet inspection in China, or congestion control in Russia.

• Countries still in the process of deciding, such as Australia or Uruguay.

As a consequence, in a global Internet, some ISPs are allowed to differentiate service for their users but also of traffic of CPs originated or going through their country. This difference of rules could end up with differentiated services for users located in a foreign and neutral country, even if neutrality is imposed and applied there. This issue is particularly exacerbated by the recent decision, in 2017, of the authorities in the USA to repeal neutrality. The USA being the origin or intermediate of an important part of worldwide traffic, it seems to us that studying the relations between countries applying different neutrality policies is becoming particularly important.

Our goal is therefore to investigate what it implies on all actors (ISPs, CPs, users) to have both neutral and non-neutral countries. To start, we will limit ourselves in this paper to two interacting countries. The actors we consider in each country are CPs, potentially deciding between classes of service for their traffic, ISPs deciding connection prices for given qualities of service (QoS) offered to CPs, and users with demand level depending on the QoS they experience. The interactions between countries come from users requesting traffic to CPs in their own and also in foreign countries, and from traffic having to go through ISPs in the two countries, potentially applying different neutrality policies. Non-cooperative game theory [4, 12] is used to analyse the interactions of selfish actors. The question we would like to answer is: Is there a “winner” with such a heterogeneous situation? We wish to compare the output with the situation of a fully neutral Internet.

The literature on modeling and analysis of network neutrality through game theory has been extensive (see among others [1,2,3, 6, 9, 10] and references therein), trying to answer various questions, but to our knowledge no work has been dealing with the impact of interactions between countries applying different rules. Again, given the current tendency of countries to evolve in different political directions, this is becoming an issue of primary interest.

The results provided in this paper highlight how the specifics of net neutrality regulation relaxation can affect all actors, including those in the region where neutrality remains enforced. Our numerical results show in particular that if the non-neutral ISP still has to offer the same quality as the neutral ISP (in addition to an improved quality), then all actors are likely to benefit from the relaxation, while only the non-neutral ISP would benefit if it is given more freedom (even its hosted CPs would prefer the all-neutral situation).

The remainder of the paper is organized as follows. Section 2 introduces our two-zones, several-CP model with the available strategies and utility functions for all actors. We then investigate in Sect. 3 the case when the non-neutral ISP can only offer a high-quality service in addition to the basic one (the one in the neutral zone). In Sect. 4 we analyze numerically, on the same instance, what happens when the regulator imposes a given ratio between the low-quality and high-quality services offered by the non-neutral ISP, while the neutral ISP fixed its quality to maximize its revenue. We provide conclusions and suggest directions for future work in Sect. 5.

2 Model

2.1 Topology and Actors

The topology we study in this paper is described in Fig. 1. We consider two geographic areas, with a single ISP and several CPs in each one; the set of CPs in the neutral (resp., non-neutral) area is denoted by (resp., ). We assume a peering relationship between the two ISPs. We will consider that net neutrality is enforced in one area, while neutrality constraints will be relaxed in the other one. The neutral area may correspond to the European Union and the non-neutral one to the United States, after net neutrality regulations have been repealed.

2.2 Service Qualities and Prices

Let $q_N$ be the performance level offered by the neutral ISP, and assume that the non-neutral ISP can offer two different quality levels, namely $q_L$ and $q_H$ (with $q_L>q_H$). We assume $q_N,q_L,q_H\in [0,1]$. Without loss of generality, the higher the value of q, the better the performance. The per-unit-of-volume prices (paid by CPs to their host ISP) are $p_N$ (in the neutral area), and $p_L$ and $p_H$ for performance levels $q_L$ and $q_H$, respectively, in the non-neutral area.

Fig. 1.

Topology

Prices will be assumed fixed, imposed by the regulator(s) or decided through competition: we only focus on quality levels as decision variables for ISPs.

2.3 CP Demands and Utilities

CPs also have a per-unit-of-volume gain (from advertisement) $a_{\ell }$ for CP $\ell$. Each CP has a volume demand coming from its own area, plus another one coming from the other area, both depending on the offered performance q (defined just after). Let $D_i^j(q)$ be the demand of content of CP j from customers in area i. We will for example later consider the standard linear expressions $D_i^j(q)=\beta _i^j +\alpha _i^j q.$ Users in area i and looking at CP j located in that same area just need to use the local ISP. Thus $q=q_i$, the performance at ISP i chosen (if a choice is available) by that CP. On the other hand, when accessing the CP j from the other area, traffic goes through both ISPs, and overall performance is assumed to be the product of the two ISP qualities , so that a null quality at a network leads to a null quality along the path, and a perfect quality at a node (that is or ) reduces the overall quality to the quality of the other node along the path.

The utility of CP $\ell$ in the non-neutral area is the difference between its advertising revenue and payments to its ISP, both of which are proportional to its total demand:

$\ell$ in non-neutral area $\Rightarrow$

where $k_{\ell }\in \{L,H\}$ is the performance level chosen by the CP.

Similarly, the utility of a CP $\ell$ in the neutral area is given by

$\ell$ in neutral area $\Rightarrow$

with decision $k_{\ell }\in \{ L,H\}$ representing the quality that the CP has selected for its flows reaching the non-neutral zone. Note that we assume that if the CP selects the high quality, it then has to pay an extra $p_H-p_L$ per demand unit to the non-neutral ISP.

2.4 ISP Utilities

The revenues of ISPs are made of the gain from the volume of data flowing through it, minus the cost for maintaining a network with the given performance level. We therefore get, for the ISP in the neutral area,

and for the ISP in the non-neutral area we obtain

where and are the cost functions. We assume that the cost borne by an ISP is made of the demand level at the quality times the (unit) cost $c_i(q_i)$ ($i\in \{N,H,L\}$) to provide this level:

Remark that total demand through a network is composed of the total demand of subscribers to this network, but also of the demand of users of the other network to CPs in this local network.

The cost functions and will be assumed increasing convex, with value 0 at 0 and $+\infty$ at 1. The typical example we will use is with some conversion rate $\alpha$.

2.5 Analysis of the Interactions

Each actor takes its own decision, but not at the same time scale. We end up with a Stackelberg game, where:

• 1. ISPs play a game on performance levels, the one in the neutral area choosing $q_N$, the one in the differentiated area choosing $q_L$ and $q_H$ (prices being fixed, as described previously);

• 2. The CPs decide the class of service in the differentiated zone;

• 3. Demand is computed depending on those strategies.

The game is played by backward induction, i.e., each decision maker is assumed to be able to anticipate the outcome of the later stages when selecting an action.

2.6 Different Ways of Relaxing Neutrality

We take as a reference basis the situation where both areas are neutral, and investigate what happens when one region allows some non-neutral behavior from its ISP. To avoid extremely unfair behaviors, we assume that there remains some (relaxed) regulation regarding the non-neutral ISP’s actions. For example, we may assume that the ISP may only offer an improved service in addition to that of the reference situation (our first scenario), or that the regulator imposes a fixed quality difference between the high- and low-quality services. Of course, other rules can be imagined, but we think those two are sufficiently simple and realistic to be worth considering.

3 A First Scenario: Opening to Non-neutrality in a Zone from an All-Neutral Situation

In this section, we investigate some possible outcomes when, from a situation where both zones are neutral and provide the same quality $q_N$, the possibility of creating fast-lanes is opened in one zone while the minimum quality should remain $q_N$. This could for example correspond to the situation in the United States, where previous neutrality recommendations have been repealed.

With our previous notations, this corresponds to $q_N$ being fixed, and the new decision variable being $q_H$.

The other parameters are also assumed fixed, with values given below.

Those values have been chosen quite arbitrarily, but so that the players’ strategies have an impact. We nevertheless think they can be realistic.

The sets of CPs are and . We consider linear demand functions for CPs, given in Table 1. In particular, we are considering pairs of CPs, i.e., CPs that have the same demand functions in both regions and only differ by being attached to the neutral or the non-neutral ISP. This will help analyze whether a given CP prefers being hosted by a neutral or a non-neutral ISP. We also consider three kinds of CPs, with one kind (CPs and ) equally of interest in both areas, and each of the two other types mainly of interest for users in one area: CPs and rather target users in the non-neutral area, while CPs and produce content that interests mostly users in the neutral area.

Table 1. CP demand functions and advertisement revenue factors.

Fig. 2.

CP utilities versus $q_H$, for $q_L=q_N= 0.5$. The situation when $q_H= 0.5$ corresponds to the all-neutral case. The marks highlight the values after which a CP selects $q_H$ rather than $q_L$.

Fig. 3.

ISP revenues versus $q_H$ when $q_L=q_N$.

We display in Fig. 2 the utilities of all CPs when $q_H$ varies, with $q_N=q_L$ fixed. We first remark that, as expected, when $q_H=q_N$ each CP is better off being hosted in the region where most of its demand lies:

• being equally attractive in both regions, CPs and get the same utility, suggesting that a “virtual CP1” considering where to locate its service would be indifferent;

• similarly, a “virtual CP2” (resp., CP3) asking the same question would compare the utility of our CPs and (resp., CPs and ) and prefer to be in the situation of CP (resp., ). This was to be expected, since CPs will seek to minimize the multiplicative impact of qualities on the demand from the alien area.

As $q_H$ increases, for our parameter values each CP opts for the improved quality after a certain point (indicated in the figure). Note that the values for this point conform to the intuition: a CP will switch sooner to high-quality if the impact on its demand is larger, i.e., if its target market is in the non-neutral zone and if the CP is itself hosted in the non-neutral zone. For the same reason, the improved quality $q_H$ favors more the CPs located in the non-neutral area than their neutral-area-located counterparts, due to the multiplicative term in the demand from the area remote to the CP. When $q_H$ becomes high, all CPs – including the one mostly targeting users in the neutral area, our “virtual CP3” – prefer being hosted in the non-neutral area than the neutral one.

The net revenues (utilities) of both ISPs are also plotted in Fig. 3, illustrating how the non-neutral ISP could choose its high-quality level $q_H$. We notice some discontinuities, corresponding to CPs switching from low-quality $q_L$ to high-quality $q_H$. Those switches first have no noticeable impact on the neutral ISP, but then a positive impact since demand increases with $q_H$; however we do not consider here what would happen if CPs decide to switch regions (as we pointed out above, the non-neutral ISP becomes more attractive to CPs as $q_H$ increases).

Fig. 4.

Best-response $q_H(q_N)$ when $q_L=q_N$ (no value is shown when offering a high-quality service is not beneficial, i.e., when the non-neutral ISP prefers offering only $q_N$).

Fig. 5.

CP utilities (revenues) versus $q_N=q_L$, with best-response $q_H$ from the non-neutral ISP.

Finally, in Figs. 4, 5 and 6 we vary the common value of $q_H$ and $q_L$, and consider that the non-neutral ISP selects the quality $q_H$ maximizing its net revenue: that best-response quality $q_H$ is plotted in Fig. 4, while Figs. 5 and 6 show the utilities of the actors. As could be expected, when the “low-quality” levels $q_N=q_L$ is high enough, a high-quality level does not make a significant enough difference, and is therefore not implemented by the non-neutral ISP. With our parameter values, this happens when $q_N$ exceeds 0.71. Note also that before that point, the optimal $q_L$ is not monotone in $q_N$: it first slowly decreases for low values of $q_N$, but for $q_N$ high enough there seems to be a need to maintain $q_H$ a certain level above $q_N$ to keep attracting CPs, i.e., justifying the price difference, as illustrated in Fig. 3 when $q_N=0.65$.

The impact on each individual CP is shown in Fig. 5: when the neutral (basic) quality is low, all CPs prefer being hosted by the non-neutral ISP since they can benefit from its high-quality level and the corresponding demands. However, when that basic quality increases, the gain from the high-level quality $q_H$ is less significant, and after some threshold (around 0.35 for our example) the CP mainly targeting users in the neutral zone prefers to be hosted in that zone. The CP targeting the non-neutral area always prefers being hosted by the non-neutral ISP, while the CP equally demanded in both regions prefers the non-neutral one until the non-neutral ISP non longer offers service differentiation.

We also display in Fig. 6 the revenues of both ISPs, and the cumulative revenues of the CPs in each zone. This allows to see what an advantage it is for an ISP to be allowed to offer an improved service; we also see the two regimes (when it is beneficial for the non-neutral ISP to offer $q_H>q_N$, and when it is not).

Fig. 6.

ISP and cumulative CP utilities (revenues) versus $q_N$, with best-response $q_H$.

Finally, in Figs. 7 and 8 we compare the all-neutral situation (when the same quality $q_N$ is offered in both areas) to the partially neutral one studied here, where a high-quality $q_H$ can be offered in the non-neutral region. We observe that for our parameters, relaxing the neutrality constraint benefits to all actors, except (for a small range of values of $q_N$) the neutral ISP. All CPs benefit from that relaxation, in particular those hosted by the non-neutral ISP. Note however, again, that we did not consider CP mobility in our model (CPs that would switch ISPs).

Fig. 7.

CP utilities (revenues) differences when switching from an all-neutral situation to a non-neutral ISP optimizing $q_H$, versus $q_N=q_L$.

Fig. 8.

ISP utilities (revenues) differences when switching from an all-neutral situation to a non-neutral ISP optimizing $q_H$, versus $q_N=q_L$.

4 A Second Scenario: Quality Game Between ISPs, with a Fixed $q_H/q_L$ Ratio in the Non-neutral Region

We now consider a different scenario, where both ISPs play a non-cooperative game on their qualities–the neutral ISP chooses $q_N$ and the non-neutral ISP chooses $q_L$ and $q_H$–but a fixed ratio between $q_H$ and $q_L$ is imposed to the non-neutral ISP, to avoid excessive differences between basic and improved services. For the numerical investigations, we consider the same CPs with their specificities (location and demand functions) as in the previous scenario. They are given in Table 1.

Fig. 9.

CP utilities versus $q_H$, for $q_N= 0.5$ and the ratio $q_H/q_L$ fixed to 2. The marks show after which value of $q_H$ the CPs prefer $q_H$ over $q_L$.

Fig. 10.

ISP revenues versus $q_H$, for $q_N= 0.5$ and the ratio $q_H/q_L$ fixed to 2.

When the quality $q_N$ is fixed and the non-neutral ISP varies $q_H$ (and $q_L$ to maintain the quality ratio), the utilities of the CPs are plotted in Fig. 9. We observe the same trend as with the previous scenario: the CPs hosted in the non-neutral region prefer the “higher quality” $q_H$ over $q_L$ sooner (i.e., for smaller values of $q_H$) than their counterparts hosted in the neutral region. Also, CPs whose main demand lies in the non-neutral region switch sooner to $q_H$. Another thing worth noting is that for all CPs, the switching point is below the quality value $q_N$ in the neutral network: here even if $q_H$ cannot be said to be high quality, CPs still choose it to avoid the worse quality $q_L$.

The net revenues (utilities) of both ISPs are also plotted in Fig. 10, illustrating how the non-neutral ISP could choose its high-quality level $q_H$. For our numerical values, the non-neutral ISP should choose a high quality around 0.8, and therefore a basic quality 0.4.

Fig. 11.

ISP best-responses when the ratio $q_H/q_L$ is fixed to 2.

Fig. 12.

Equilibrium qualities when the ratio $q_H/q_L$ varies.

However, in this scenario we do not consider $q_N$ fixed but rather determined through a non-cooperative game played between ISPs. To analyze that game, we plot the ISP best-responses ($q_H$ versus $q_N$, since $q_L$ is directly determined by $q_H$) in Fig. 11, which exhibits continuous best-response that intersect at only one point, a (stable) Nash equilibrium. Note that the best-response $q_H$ is above $q_N$, and seems to equal $q_N$ when $q_N$ is large enough. Here at the equilibrium we have $q_H>q_N$.

We then vary in Fig. 12 the regulated ratio $q_H/q_L$, and plot the equilibrium values of the qualities. The resulting revenues of ISPs and CPs are displayed in Figs. 13 and 14. The figures highlight two regimes:

Fig. 13.

ISP net revenues at equilibrium when the ratio $q_H/q_L$ varies.

Fig. 14.

CP net revenues at equilibrium when the ratio $q_H/q_L$ varies.

• when the ratio $q_H/q_L$ is low, the non-neutral ISP cannot set a significantly higher quality $q_H$ to justify the price difference, and all CPs prefer the basic quality $q_L$. The results therefore do not depend on the specific value of the regulated ratio, provided it is low enough. Note that the equilibrium quality $q_L$, the one all CPs choose, is slightly below the quality of the neutral ISP, even though their price is the same. That regime is actually equivalent to the all-neutral equilibrium (when the ratio is 1, i.e., no differentiation is allowed); we have the same qualities (except $q_H$, that is not chosen by any CP) and the same utilities for stakeholders. Therefore, such a limited relaxation of neutrality constraints has no impact.

• When the regulated ratio $q_H/q_L$ is large enough, the price difference is justified by the quality difference, hence all CPs select that high-quality. Again, as a result the equilibrium qualities do not depend on the regulated ratio in that regime. Note that in this regime, the chosen quality in the non-neutral zone $q_H$ is strictly above the one in the neutral zone. It is also worth noting that both ISPs–even the neutral one–prefer this regime over the all-neutral situation, while it is the opposite for CPs: all of them were better off in the all-neutral setting than in this relaxed scenario, the most affected being the CPs hosted in the non-neutral zone.

Between those two regimes, there is a small range of values for the regulated ratio, where some CPs would select $q_L$ and others $q_H$. The impacts on the actors in that limited range are less clear.

5 Conclusions

Changing the net neutrality rules in a part of the Internet may affect actors in other parts; in this paper we have focused on a simple scenario with two domains having different regulations, and have investigated the resulting decisions from ISPs (in terms of offered qualities) and the consequences on content providers.

The specifics of the new rules can lead to very different outcomes. With the same parameter values, our numerical analysis has for example shown that, when a non-neutral ISP can offer two qualities $q_H$ and $q_L$ and a neutral ISP only one quality $q_N$:

• if we impose $q_L=q_N$, i.e., non-neutrality is only allowed to offer improved service, then all stakeholders–in particular the non-neutral ISP and its hosted CPs, but also the other ones to a lesser extent–would prefer a relaxation of neutrality rules;

• if a given ratio $q_H/q_L$ is imposed, then either nothing is changed with respect to the neutral situation, or all CPs are worse off.

Those results suggest that regulatory decisions about relaxing net neutrality rules should be made with great care, taking into account the whole ecosystem that is affected.

The setting described in this paper opens several perspectives for future work. A first direction would be to develop the analytical study of the model proposed here rather than investigate numerical examples. Also, one can imagine other versions of regulation to constrain the non-neutral ISP in other ways, with the objective to favor user welfare, CP innovation, or fairness among users or among CPs.