Innovational Complementarities and Network Neutrality Johannes M. Bauer and Günter Knieps Michigan State University and University of Freiburg TPRC 43.

Innovational Complementarities and Network Neutrality

Johannes M. Bauer and Günter KniepsMichigan State University and University of Freiburg

TPRC 43 Arlington, VA, 24-27 September 2015

2

Motivation• FCC net neutrality order and much of policy

debate based on assumption that Internet innovation is edge-driven

• Most economic models of the effects of net neutrality regulation rely on a specific, well-understood but narrow, framework (M/M/1) to model congestion

• Paper seeks to broaden analysis to a more general model of interdependent innovation

3

Diversity and heterogeneity of uses

Source: Sandvine 2015

4

Continuous network innovation(Download capacity for DSL, cable, and mobile 1988-2015)

Source: Bauer & Latzer, forthcoming

5

Interdependent innovation• Innovation– Combination and re-combination of knowledge– Evolutionary search process

• Drivers– Opportunities (“adjacent possible”)– Appropriability of rewards– Capabilities

• In the ICT system innovation conditions at each layers enable and constrain the innovation conditions at the other layers

6

Framing Internet innovations

• Modular core and edge innovations– End-to-end does not per se oppose active traffic

management but suggests that preserving low-cost options to innovate on the edges of the network has substantial value

– This value is unlocked by keeping the core network services and functions simple and cheap

• Network ownership versus ownership at the edge– Considerable advantages in a network architecture in

which innovators at the edge do not intervene in the competencies of network operators and vice versa

7

M/M/1 queuing models• Based on Poisson process of arrival rate of service requests• Average waiting time in the priority class differs from

average waiting time in the best effort class• Analyzes average waiting time depending on network traffic

and transmission capacity• Given the stochastic nature of the Poisson process,

deterministic traffic quality guaranties of maximal end-to-end response time of any data packet in the top priority class are beyond the M/M/1 framework

• Additional investments increase transmission capacity and thereby increase average service

8

All-IP networks • The narrow focus on end-to-end response time of

content delivery can and needs to be extended to encompass real time jitter-sensitive applications.

• To analyze the full innovation potential at the edge and within all-IP networks and their interplay/complementarities demands a more general approach toward traffic quality.

• It requires taking into account not only stochastic traffic quality (e.g. average expected response time) but also deterministic traffic quality in determining maximum response times (D) and maximum jitter (J).

9

Complementary GPTs• Innovations within the Internet are not only driven by

applications but can also be stimulated by developments at the network and traffic layers.

• The all-IP infrastructure and Generalized DiffServ architecture function as General Purpose Technologies (GPTs) for applications and services.

• It is important that the GPTs on the broadband infrastructure level and on the traffic architecture level are open for innovative evolutions.

• Mutual feedback effects between applications and network/traffic layers.

10

Generalized DiffServ architecture• Allows implementation of a variety of multipurpose traffic

architectures with deterministic and stochastic traffic quality guarantees by creating different traffic classes that can support time- and non-time-sensitive applications.

• Traffic quality parameters are not only limited to mean, statistical or probabilistic end-to-end response times but also manage worst case analysis of the network behavior.

• Maximum response time guarantee as well as active jitter management for real time applications can be provided.

11

An open innovation space• The Generalized DiffServ architecture contains

frameworks and building blocks for a variety of transmission architectures enabling the organization of various traffic class hierarchies.

• Basic characteristics of each entrepreneurial selection of the Generalized DiffServ architecture– Application-blindness of the traffic network– Active traffic management– Market driven network neutrality

12

End-to-end connectivity and market-driven network neutrality

• QoS differentiation and traffic class pricing in the Generalized DiffServ model do not incentivize ad hoc discrimination of specific applications.

• Instead, market-driven network neutrality is realized, where only opportunity costs of traffic qualities are relevant for pricing, irrespective of the specific application.

• Multipurpose traffic allocation rules for the total traffic, having impact on all users rather than a particular subset of users.

• Transmission capacities are shared among different traffic classes with monotonic declining traffic quality.

13

Complementarity of edge innovations

• All IP-infrastructure and Generalized DiffServ architecture function as GPTs for application services. – A large and open set of application services can be provided.– E.g., e-mail services, content delivery as well as time-

sensitive protocol session dependent, interactive services.

• In contrast, vis-à-vis the network application services at the edge do not have the characteristics of a GPT. – E.g., Internet message format standards and the simple mail

transport protocol (SMTP), real-time transport protocol for transmission of real-time data and providing QoS feedback.

14

Implications for net neutrality• Overarching framework for the discussion

– The net neutrality debate in general and the specific policies adopted in the U.S. use an overly simplified innovation concept.

– The underlying innovation model privileges innovations at the application layer but infrastructure innovations are also a driver of innovation.

• Implications for net neutrality research– Analytical models rely heavily on M/M/1 queuing model. – Understanding innovational complementarities will require more general

congestion models.

• Implications for policy design– Innovations at the application layer and at the network layer flourish

under different regulatory conditions that require balancing a trade-off.– Improperly designed net neutrality policies may bias innovation efforts in

favor of services or infrastructure resulting in lower overall innovation.

15

Recap• Innovation is an evolutionary search process of combination,

recombination and selection. • A large variety of innovations at the edge could evolve based on

TCP best effort transmission protocols. • Growing diversity and heterogeneity of applications and services

implies that network differentiation will become a more important precondition for the vibrancy of the interdependent innovation system.

• In technologically dynamic industries with asymmetrically distributed knowledge it is important to allow entrepreneurial choices of for-profit and non-profit actors to experiment freely.

• Net neutrality policy as currently specified limits the technological and economic space over which such experiments can take place.