The Public Option: A non-regulatory alternative to Network Neutrality - - PowerPoint PPT Presentation

The Public Option: A non-regulatory alternative to Network Neutrality

Richard Ma

Advanced Digital Sciences Center, Illinois at Singapore School of Computing, National University of Singapore

Joint work with Vishal Misra (Columbia University)

ACM CoNEXT 2011

The Internet Landscape

 Internet Service Providers (ISPs)  Internet Content Providers (CPs)  Regulatory Authorities  Users/Consumers

Network Neutrality (NN)

Better! Happy?

Paid Prioritization (PP)

Happier?

Highlights

 A more realistic equilibrium model of

content traffic, based on

 User demand for content  System protocol/mechanism

 Game theoretic analysis on user utility

under different ISP market structures:

 Monopoly, Duopoly & Oligopoly

 Regulatory implications for all scenarios

and the notion of a Public Option

𝝂

 𝜈: capacity of a single access (eyeball) ISP  𝑁: # of users of the ISP (# of active users)  𝒪: set of all content providers (CPs)  𝜇𝑗: throughput rate of CP 𝑗 ∈ 𝒪

Three-party model (𝑁, 𝜈, 𝒪)

⋮

𝑵

⋮ 𝒪 𝝁𝒋

User-side: 3 Demand Factors

 Unconstrained throughput 𝜄𝑗

 Upper-bound, achieved under unlimited capacity  E.g. 5Mbps for Netflix

 Popularity of the content 𝛽𝑗

 Google has a larger user base than other CPs.

 Demand function of the content 𝑒𝑗(𝜄𝑗)

 Percentage of users still being active under the

achievable throughput 𝜄𝑗 ≤ 𝜄𝑗

Unconstrained Throughput 𝜇𝑗

User size 𝑵(= 𝟐𝟏) (Max) Throughput 𝜾 𝒋(= 𝟖𝑳𝒄𝒒𝒕) Content unconstrained throughput 𝝁 𝒋 = 𝜷𝒋𝑵𝜾 𝒋(= 𝟓𝟑𝑳𝒄𝒒𝒕) Content popularity 𝜷𝒋(= 𝟕𝟏%)

Demand Function 𝒆𝒋 𝜾𝒋

achievable throughput

𝜾𝒋 demanding # of users 𝜷𝒋𝑵𝒆𝒋 𝜾𝒋

𝜷𝒋𝑵

𝜾 𝒋

 Assumption 1: 𝑒𝑗 𝜄𝑗 is continuous and

non-decreasing in 𝜄𝑗 with 𝑒𝑗 𝜄𝑗 = 1.

 More sensitive to throughput  Throughput of CP i:

𝝁𝒋 𝜾𝒋 = 𝜷𝒋𝑵𝒆𝒋 𝜾𝒋 𝜾𝒋

Demand Function 𝒆𝒋 𝜾𝒋

achievable throughput

𝜾𝒋 demanding # of users 𝜷𝒋𝑵𝒆𝒋 𝜾𝒋

𝜷𝒋𝑵

𝜾 𝒋

System Side: Rate Allocation

 Axiom 1 (Throughput upper-bound)

𝜄𝑗 ≤ 𝜄 𝑗

 Axiom 2 (Work-conserving)

𝜇𝒪 = 𝜇𝑗

𝑗∈𝒪

= min 𝜈, 𝜇 𝑗

𝑗∈𝒪

 Axiom 3 (Monotonicity)

𝜄𝑗 𝑁, 𝜈2, 𝒪 ≥ 𝜄𝑗 𝑁, 𝜈1, 𝒪 ∀ 𝜈2 ≥ 𝜈1

Uniqueness of Rate Equilibrium

 Theorem (Uniqueness): A system (𝑁, 𝜈, 𝒪)

has a unique equilibrium {𝜄𝑗 ∶ 𝑗 ∈ 𝒪} (and therefore {𝜇𝑗 ∶ 𝑗 ∈ 𝒪}) under Assumption 1 and Axiom 1, 2 and 3. User demand: {𝜄𝑗} → {𝑒𝑗} Rate allocation: μ, 𝑒𝑗 → {𝜄𝑗}  Rate equalibrium: {𝜄𝑗

∗}, {𝑒𝑗 ∗}

ISP Paid Prioritization

ISP Payoff: 𝑑 𝜇𝑗

𝑗∈𝒬

= 𝑑𝜇𝒬

$𝒅/unit traffic $𝟏

Premium Class Ordinary Class Capacity Charge

𝝀𝝂 (𝟐 − 𝝀)𝝂 𝑵, 𝝀𝝂, 𝓠 𝑵, 𝟐 − 𝝀 𝝂, 𝓟

Monopolistic Analysis

 Players: monopoly ISP 𝐽 and the set of CPs 𝒪  A Two-stage Game Model 𝑁, 𝜈, 𝒪, 𝐽

 1st stage, ISP chooses 𝑡𝐽 = (𝜆, 𝑑) announces 𝑡𝐽.  2nd stage, CPs simultaneously choose service

classes reach a joint decision 𝑡𝒪 = (𝒫, 𝒬).  Outcome (two subsystems):

 𝑁, 𝜆𝜈, 𝒬 : set 𝒬 (of CPs) share capacity 𝜆𝜈  𝑁, 1 − 𝜆 𝜈, 𝒫 : set 𝒫 share capacity 1 − 𝜆 𝜈

Utilities (Surplus)

 ISP Surplus: 𝐽𝑇 = 𝑑

𝜇𝑗

𝑗∈𝒬

= 𝑑𝜇𝒬;

 Consumer Surplus: 𝐷𝑇 =

𝜚𝑗𝜇𝑗

𝑗∈𝒪

 𝜚𝑗 : per unit traffic value to the users

 Content Provider:

 𝑤𝑗 : per unit traffic profit of CP 𝑗

𝑣𝑗 𝜇𝑗 = 𝑤𝑗𝜇𝑗 𝑤𝑗 − 𝑑 𝜇𝑗 if 𝑗 ∈ 𝒫, if 𝑗 ∈ 𝒬.

Type of Content

Value to users 𝝔𝒋 Profitability of CP 𝒘𝒋

Monopolistic Analysis

 Players: monopoly ISP 𝐽 and the set of CPs 𝒪  A Two-stage Game Model 𝑁, 𝜈, 𝒪, 𝐽

 1st stage, ISP chooses 𝑡𝐽 = (𝜆, 𝑑) announces 𝑡𝐽.  2nd stage, CPs simultaneously choose service

classes reach a joint decision 𝑡𝒪 = (𝒫, 𝒬).  Theorem: Given a fixed charge 𝑑, strategy

𝑡𝐽 = (𝜆, 𝑑) is dominated by 𝑡𝐽

′ = (1, 𝑑).

• The monopoly ISP has incentive to allocate

all capacity for the premium service class.

Utility Comparison: Φ vs 𝛺

Ψ = 𝐽𝑇/𝑁 𝜉 = 𝜈/𝑁 Φ = 𝐷𝑇/𝑁

Regulatory Implications

 Ordinary service can be made “damaged

goods”, which hurts the user utility.

• Implication: ISP should not be allowed to

use non-work-conserving policies (𝜆 cannot be too large).

 Should we allow the ISP to charge an

arbitrarily high price 𝑑?

High price 𝑑 is good when

Value to users 𝝔𝒋 Profitability of CP 𝒘𝒋

High price 𝑑 is bad when

Value to users 𝝔𝒋 Profitability of CP 𝒘𝒋

Oligopolistic Analysis

 A Two-stage Game Model 𝑁, 𝜈, 𝒪, ℐ

 1st stage: for each ISP 𝐽 ∈ ℐ chooses 𝑡𝐽 = (𝜆𝐽, 𝑑𝐽)

simultanously.

 2nd stage: at each ISP 𝐽 ∈ ℐ, CPs choose service

classes with 𝑡𝒪

𝐽 = (𝒫𝐽, 𝒬𝐽)

 Difference with monopolistic scenarios:

 Users move among ISPs until the per user utility

Φ𝐽 is the same, which determines the market share of the ISPs

 ISPs try to maximize their market share.

Duopolistic Analysis

𝓠 𝓟 𝓞

ISP 𝑱 with 𝒕𝑱 = (𝝀, 𝒅) ISP 𝑲 with 𝒕𝑲 = (𝟏, 𝟏)

Duopolistic Analysis: Results

 Theorem: In the duopolistic game, where an

ISP 𝐾 is a Public Option, i.e. 𝑡𝐾 = (0, 0), if 𝑡𝐽 maximizes the non-neutral ISP 𝐽’s market share, 𝑡𝐽 also maximizes user utility.

• Regulatory implication for monopoly cases:

Oligopolistic Analysis: Results

 Theorem: Under any strategy profile 𝑡−𝐽, if

𝑡𝐽 is a best-response to 𝑡−𝐽 that maximizes market share, then 𝑡𝐽 is an 𝜗–best-response for the per user utility Φ.

• The Nash equilibrium of market share is an

𝜗-Nash equilibrium of user utility.

• Oligopolistic scenarios:

Regulatory Preference

Monopoly Oligopoly

ISP market structure User Utility