May 4, 2012Bell Labs, Crawford Hill Time-Dependent Pricing of Mobile Data Soumya Sen Princeton University Joint work with: Sangtae Ha, Carlee-Joe Wong,

May 4, 2012 Bell Labs, Crawford Hill

Time-Dependent Pricing of Mobile Data

Soumya Sen

Princeton University

Joint work with:

Sangtae Ha, Carlee-Joe Wong, Mung Chiang

I. Motivation

Wireless Internet Usage Trends

Mobile data growing at 78% annually

Driving Forces

Mobile Video

CloudSync

Data-hungryApps

High-resDevices

A PerfectStorm

Ultra-Heavy Tail

ISP cost structure’s fundamental problem

But Not Heavy All the Time

Large Peak-Valley Differential

Time Elasticity: Opportunities

Time Elasticity

Volu

me

Streamingvideos,Gaming

Texting,Weather, Finance

Email,Social

Network updates

Cloud

SoftwareDownloads

Movies & Multimedia downloads,

P2P

Opportunities

Cost Reduction

ISP’s Spectrum, Capital, Operational costs decrease with reduced peak

Time Time

Before After

Ban

dwid

th

Peak

Peak

Ban

dwid

th

Revenue Increase

Time Time

Before After

Ban

dwid

th

$50 for 5 GB $60 for 10 GB

Create win-win by increasing demand

Ban

dwid

th

II. Feasibility Study

Consumer Response

USA: Online Survey, 130 participants, 25 states

India: Face-to-face Surveys: 550 participants, 5 cities Professionals (36%), Students (36%), Self-employed (8%),

housewives (6%), unemployed (12%)

Time Elasticity: Survey Results

YouTube streaming Downloads

Many applications are time-elastic

Policy Feasibility

FCC Dec. 2010 Statement“...the importance of business

innovation to promote network investment and efficient use of networks, including measures to match price to cost, such as

usage-based pricing”

Industry Moves: US ISPs

Industry Moves: Indian ISPs

Industry Moves: African ISPs

Africa dynamic pricing

Current Practices

✤ Flat Rate, throttling heavy-users

✤ Usage-based Pricing

✤ David Clark, ’95: “The fundamental problem with simple usage fees is that they impose usage costs on users regardless of whether the network is congested or not.”

✤ Dynamic Pricing

✤ MacKie-Mason, ’95: “We argue that a feedback signal in the form of a variable price for network service is a workable tool to aid network operators in controlling Internet traffic. We suggest that these prices should vary dynamically based on the current utilization of network resources.”

History of Pricing Research

Other Markets: Electricity

“Day Ahead” Pricing

* Sen, et al., “A Survey of Broadband Data Pricing: Past Proposals, Current Plans, and Future Trends”, 2012.

III. Challenges

Key Questions

✤ Optimized Price Computation?

✤ Correct Incentives? TUBE Theory

✤ Practical Economic Modeling

✤ System Design Issues

✤ How to assess TDP benefits?

✤ Will real customers respond?TUBE Trial

TUBE System

IV. TUBE Technology

Time Dependent Pricing (TDP)

Large scale ISP cost optimization, taking user reaction into account

ISP’s Optimization Problem

Cost of overshooting

capacity

Cost of rewards

Estimating Waiting Function

Economic modeling

reward

patience indexdelay

waiting function

Patience Index: Initialization

TDP: Shifting Peak to Valley

TDP Performance

V. TUBE Princeton Trial(May 2011-January 2012)

Princeton Trial: Money Flow

• 50 AT&T participants : 27 iPhones, 23 iPads• Faculty, staff, and students• 14 Academic units

TUBE App: Information Screens

TUBE App: Scheduling Screens

VI. Princeton Trial Results

Usage Statistics

✤ How much bandwidth participants use? – ‘Heavy tailed’

✤ Which applications use the most bandwidth? – Video streaming

July-September, 2011

20%75%

Price Sensitivity

✤ Do users wait to use mobile data in return for a monetary discount?

✤ Average usage decrease in high-price periods relative to the changes in low-price periods (iPads: -10% in high-price, 15% in low-price periods)

October 2011

Notification Effectiveness

✤ Do notifications impact usage?

✤ 80-90% of users decrease or did not change their usage after the 1st notification

✤ For all subsequent notifications, about 60-80% of the active users decrease their usage, while the others remained price-insensitive.

iPads iPhones

Psychological Factors

✤ Do users respond more to the numerical values of TDP prices or to the color of the price indicator bar on the home screen?

Period Type 1 and 3 Period Type 1 and 2

Optimized TDP Impact

✤ Does the peak usage decrease with time-dependent pricing? And does this decrease come at the expense of an overall decrease in usage?

✤ Optimized TDP reduce the peak-to-average ratio (max reduction: 30%)

✤ Overall usage increase with TDP (demand gain in valley periods)

Peak-to-Average Ratio Peak Usage Volume

Impact on Web Ecosystem

✤ Does the application usage distribution change due to TDP?

✤ People are motivated to use more bandwidth during low-price periods, “valley filling”.

VII. Post-Trial Survey

Viability

✤ Will you be able to decide on “when” to use?

✤ “I think it's a great idea, ..the iPads would say, 'If you wait a half an hour, you can have...' I thought that was incredibly useful. And I would be able to make that decision.”

✤ Are there apps for which you usually wait?

✤ “[I]f I'm out in my car and I needed it for GPS, I wouldn't care how much money I'm spending… if I just wanted to be on a social network or check my email, I would certainly wait.”

Usefulness

✤ What are your main concerns with TDP?

✤ “If it's predictable, yes, I think so, because let's say I know that definitely everyday from 9 to 10 it's less, then I can plan a little bit.”

✤ Was the color-coded notification bar useful to you?

✤ “I group the colors I would see if it's a good color for me... because I couldn't always figure out what it meant in terms of the dollar amount and translate that into how much I was using”

Opinions

✤ Were you tempted to use more data when the discounts were higher?

✤ “[laughs] Kind of! But that also goes toward my personality of if it's on sale I must buy it!”

✤ Will TDP adversely affect high-bandwidth app developers?

✤ “I don't think this will result in those kinds of applications being developed less, and I think that's because you're giving users the option”

Related Publications:

[1] “TUBE: Time-Dependent Pricing of Mobile Data”, SIGCOMM 2012.

[2] “A Survey of Broadband Data Pricing: Past Proposals, Current Plans, and Future Trends”, under submission in ACM Computing Surveys.

Thank you

http://scenic.princeton.edu/tube/

Princeton Workshop on

Smart Broadband Pricing

[email protected]