Social Welfare and Policy Impact on Emerging ISP-Content Provider Relationships
23 Pages Posted: 27 Mar 2016 Last revised: 16 Aug 2016
Date Written: March 22, 2016
The paper develops, analyzes and numerically investigates models of the relationships between broadband access ISPs (ISP) and Content Providers (CP) in different forms, some that are emerging, and others that appear likely to emerge in the near future. The Netflix-Comcast relationship is viewed as prototypical, a likely pioneer of an emerging trend. Netflix has developed the large-scale caching system Open Connect and deployed it in collaboration with Comcast and other ISPs widely across the globe. Business relationships between Netflix and ISPs have quickly gone from being adversarial to cooperative. Also, several of Netflix’s fellow-CP and competitors, such as Amazon and Hulu, are following the same path.
Our models capture essential features of decision-making in systems engineering and investments in the emerging business relationships and regulatory environments. The systems engineering is of content delivery systems, with caching and bandwidth allocation as the main elements. Also included is the Quality of Experience (QoE) that influences potential users’ decision to subscribe to the content service, and the user’s price elasticity of demand for the service. The investment dimension is represented by the CP’s decision on the size of caches (and corresponding interconnection facilities at the CP-ISP interfaces) that it will build, the amount of bandwidth that the content service will need, and the number of subscribers to the service. Motivated by uncertainties of the strictness with which Net Neutrality regulations will be applied, we study their impact by analyzing the model for different scenarios. In like vein, the degree of cooperation that CP and ISPs can exercise in making business decisions on pricing, bandwidth allocation and cache deployment in our models is subject not only to Net Neutrality regulations but also anti-trust laws. Here too the uncertainty has persuaded us to consider alternative scenarios. The models for the different scenarios give numerical results that are insightful to compare and contrast, and they also offer pointers to model extensions and generalizations.
The CP’s content is typically stored in the cloud remote from the end user, and in our model the content desired by the end user may have to traverse two separate networks, the Internet backbone and the access, “last mile”, network. Let r_1 and r_2 respectively denote the average data rate over the access network and the Internet backbone,where r_1 >r_2. If there is nothing else, then the average data rate, which is the “effective bandwidth” or effective rate, r_eff=min(r_1,r_2 )=r_2.
Now consider the impact of (i) a cache of size S located close to the interface between the Internet and the access network, and (ii) a rate enhancement of β over the last mile for subscribers to the content service. It is not necessary for both mechanisms to be simultaneously operational. The cache will yield a “hit probability”, h , of finding the content desired by the end user in the cache. The hit probability has been extensively studied in the networking literature, and shown to be approximately proportional to logS. In the event of a “hit”, the content desired by the customer traverses only the access network, i.e., the traversal over the Internet backbone is circumvented. Hence, the combined effect of (i) and (ii) is an effective data rate, r_eff= (r_1 β) h r_2 (1-h).
This is the average data rate of content delivery enjoyed by users of the content service. Both the hit probability and the rate enhancement has the potential to significantly influence the user’s QoE. We use well-known results from the networking literature to translate the effective data rate to the user’s QoE. The QoE, the subscription price and the price elasticity of demand combine to determine the demand for the content service.
In the base case the decisions on S and β are made according to the Stackelberg model, which is well-known in economics and game theory. In our model the ISP is the “leader” and sets the unit prices for caches and extra bandwidth. Responding to the ISP’s price setting, the “follower” CP determines S and β, as well as the subscription fee for the service, to maximize its profit. Next, the ISP does its own profit maximization with respect to the prices for caches and extra bandwidth in which it incorporates knowledge of the CP’s best response policy. The mantle of the leader falls on the ISP quite reasonably since it owns the last mile infrastructure, which gives it a stronger bargaining position with the CP. The ISP’s price setting is done with enlightened self-interest, because it knows that setting the prices too high for the CP will lead ultimately to reduced demand.
The alternative scenarios that we consider cover two separate issues, namely, (i) whether or not Net Neutrality regulations allow enhancements β to the data rate for the content service over the access, last mile, network, and (ii) whether or not the CP and ISP cooperate in making decisions on cache size (S) and extra bandwidth (β). That is, for (ii), in one scenario the ISP and CP make a coordinated decision on S and β to maximize their combined profits, and in another scenario the Stackelberg leader-follower model of decision-making applies.
We report on extensive numerical investigations on the models. These are MATLAB- based, some use exhaustive search for solutions to the respective optimizations performed by the CP and ISP, and yet others are based on analytic results. For some problems we use both methods to validate the results.
Our numerical investigations demonstrate that the price-elasticity of end-user demand plays an influential role in deciding the outcomes of the models, suggesting that it would be an important consideration in deciding appropriate regulatory policies. For instance, strict enforcement of Net Neutrality regulations, which would constrain β=0, has significant impact on the usage of the service and the profits of the ISP and CP when the value of the elasticity is low. We also find that the integration of decision-making by the ISP and CP can lead to a significant increase of their combined profit over the Stackelberg game.
Keywords: Internet policy, regulatory policy, ISP-CP relationships, Stackelberg games, caching, content delivery, QoE
JEL Classification: L51, L86, L96, C70, C61
Suggested Citation: Suggested Citation