IEEE Communications Magazine • November 2012 910163-6804/12/$25.00 © 2012 IEEE

1 M. Reardon, “AT&TWi-Fi Usage Skyrockets,”Cnet News, Oct. 24,2011.

2 E. Caggianni, “GrowingData Demands are Trou-ble for Verizon, LTECapacity Nearing Limits,”Talk Android, Mar. 6,2012.

3 A. Lee, “AT&T to BeginBroadband Caps,” Huff-ington Post, May 2, 2011.

4 L. Segall, “Verizon EndsUnlimited Data Plan,”CNN Money, July 6,2011.

5 D. Goldman, “ComcastScraps Broadband Cap,Moves to Usage-basedBilling,” CNN Money,May 17, 2012.

6 J. Cryderman, “Leverag-ing Policy 2.0 and Analyt-ics,” Pipeline Magazine,vol. 9, no. 2, July 2012,http://media.pipeline.pub-spoke.com/files/issue/14/PDF/PipelineJuly2012_A4.pdf, pp. 3.

INTRODUCTION

The demand for data is surging rapidly everyyear, with the Cisco Visual Networking Index for2012 projecting an 18-fold increase in globalmobile data traffic from 2011 to 2016. Mobilevideo is predicted to be the fastest growing con-sumer mobile service, increasing from 271 millionusers in 2011 to 1.6 billion users in 2016 [1].Much of this demand growth is driven by the pop-ularity of smart devices, bandwidth-hungry appli-cations, cloud-based services, and media-rich webcontent [2]. This explosive growth in bandwidthdemand is now forcing Internet service providers(ISPs) to invest in and expand their wired andwireless capacity by acquiring additional spec-trum, deploying WiFi hotspots, and adopting newtechnologies such as fourth-generation (4G) LongTerm Evolution (LTE) and femtocells. However,even these measures are proving insufficient formeeting today’s challenges of network

congestion.1,2 Consequently, major U.S. ISPs likeAT&T,3 Verizon,4 and Comcast5 have been aban-doning the traditional unlimited “flat rate” dataplans in favor of congestion-reducing mechanismslike throttling, data caps, and tiered usage-based(metered) pricing [3]. But such moves have beenfraught with concerns for an open Internet, lead-ing to a polarizing debate on how ISPs shouldmanage their network traffic to create a sustain-able Internet ecosystem.

This debate has primarily focused on theissues of:• Who should pay the price of congestion?• What form should such pricing or network

management policies take?The former question, which has serious net-neu-trality implications [4, 5], relates to the issue ofwhether content providers should pay a part oftheir revenues to finance ISPs’ capital invest-ments in the underlying network infrastructure.The latter question, which relates more directlyto the theme of this article, centers on the fair-ness and appropriateness of current pricing poli-cies, such as application- and usage-basedpricing. App-based pricing, such as toll-free(zero-rated) access to specific applications aspracticed by Mobistar6 in Belgium, is arguablydiscriminatory and much more contentious froma net-neutrality standpoint [3]. However, usage-based pricing (or data caps with metering) iswidely viewed as an acceptable way to “matchprice to cost,” even by the U.S. Federal Commu-nications Commission (FCC) [6]. Yet the effec-tiveness of usage-based pricing (UBP) has beenquestioned by both academic and industry veter-ans due to the fact that usage fees impose costson users regardless of the actual network con-gestion in the network at any given time. Forinstance, Vinton Cerf [7] wrote on Google’sPublic Policy blog, “Network management alsoshould be narrowly tailored, with bandwidth con-straints aimed essentially at times of actual conges-tion. In the middle of the night, available capacitymay be entirely sufficient, and thus moderatingusers’ traffic may be unnecessary.” AndrewOdlyzko et al. [3] echoed a similar view in statingthat “Unless UBP contains a time-of-day billingfeature and some immediate feedback on conges-tion it is hard to imagine how it can be used as acongestion management tool.”

ABSTRACT

The tremendous growth in demand for broad-band data is forcing ISPs to use pricing as a con-gestion management tool. This changinglandscape of Internet access pricing is evidencedby the elimination of flat rate data plans in favorof usage-based pricing by major wired and wire-less operators in the US and Europe. But simpleusage-based fees suffer from the problem ofimposing costs on all users, irrespective of thenetwork congestion level at a given time. Toeffectively reduce network congestion, appropri-ate incentives must be provided to users who arewilling to time-shift their data demand from peakto off-peak periods. These pricing incentives caneither be static (e.g., two-period daytime/night-time prices) or computed dynamically (e.g., day-ahead pricing, real-time pricing). Data plans thatoffer such incentives to consumers fall under thecategory of time-dependent pricing (TDP). ManyISPs across the world are currently exploring var-ious forms of TDP to manage their traffic growth.This article first outlines the sources of today’schallenges, and then discusses current trendsfrom regulatory and technological perspectives.Finally, we review representative pricing propos-als for incentivizing the time-shifting of data.

COMMUNICATIONS NETWORK ECONOMICS

Soumya Sen, Carlee Joe-Wong, Sangtae Ha, and Mung Chiang, Princeton University

Incentivizing Time-Shifting of Data: A Survey of Time-Dependent Pricing forInternet Access

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

____________________________

________________

IEEE Communications Magazine • November 201292

Cerf and Odlyzko essentially argued thatUBP is not an appropriate pricing scheme forcongestion alleviation, as it does not address theproblem of several users simultaneously access-ing the network resources, which in turn leads tolarge demand peaks and poor network perfor-mance at those times. What ISPs can benefitfrom are smart data pricing7 (SDP) mechanisms,which can create the right level of time-varyingincentives that induce consumers to shift theirnon-critical and bandwidth-heavy applications toperiods of lower congestion. Time-dependentpricing (TDP) schemes focus on this aspect, andhence can significantly help ISPs manage theirnetworks. In fact, TDP for voice calls has beenwidely practiced in the United States by AT&Tand Verizon, with free calling on nights andweekends but caps on minutes used during theday. More dynamic versions of TDP have beenimplemented by MTN Uganda and UninorIndia, which update the prices for voice callshourly, depending on the congestion conditionsat the call’s originating location [8]. Similar TDPschemes for broadband data have been proposedrecently [2, 9] with a full system implementationand trials [9, 10]. But it should be noted thatTDP does not necessarily need to have an explic-it pricing signal to the user; it can be implement-ed as an intelligent flat rate (IFR) scheme byautomating the client side to change consumerusage behavior in response to a feedback signalfrom the network to stay within a given budget.

TDP leverages the trade-offs between con-sumers’ price sensitivity and delay tolerance(time elasticity of demand) across differentapplications to benefit both consumers and ISPs:consumers benefit from more flexibility ofchoice, while ISPs benefit from reduced costs ofcapacity provisioning. Given the level of interestthat TDP has received within both academia and

industry [2, 9, 10], and its potential as a viablepricing policy for broadband data, this articlereviews several TDP research works, with a par-ticular focus on incentive-based models for time-shifting of data. To contextualize the problem,we first provide a brief background of the keyfactors that are contributing to network conges-tion, followed by an analysis of the trends inpricing policy changes that ISPs have adopted inresponse and their regulatory implications.

KEY DRIVERS OFNETWORK CONGESTION

Technological advances as well as changing con-sumer behavior are driving the explosion in datademand. We discuss these factors next and pro-vide some additional growth statistics in Fig. 1.

BANDWIDTH-HUNGRY DEVICESThe widespread adoption of handheld devices,equipped with powerful processors, high-resolu-tion cameras, and larger displays, has made itconvenient for users to stream high-qualityvideos and exchange large volumes of data. Datafrom laptops equipped with 3G dongles and net-books with wireless high-speed data access con-tributes the most to wireless network congestion[2]. As for smartphones, Cisco projects that theaverage monthly data usage will rise from 150MB in 2011 to 2.6 GB in 2016 [1]. New featureslike Siri on the iPhone 4S, which has doubledApple users’ data consumption, are driving thisgrowth.8

CLOUD SERVICE AND M2M APPLICATIONSCloud-based services that synchronize dataacross multiple mobile devices, such as iCloud,Dropbox, and Amazon’s Cloud Drive, can be a

7 Smart Data Pricing(SDP) Forum,http://scenic.princeton.edu/SDP2012/. SDP is abroader notion than time-shifting of data. It can bea combination of ideas,including fair throttling,smart demand shaping,application- or location-dependent pricing, intelli-gent WiFi offloading,quota-aware content dis-tribution, optimized con-tent caching andforwarding, and spon-sored access.

8 J. Browning, “Apple’sSiri Doubles iPhone DataVolumes,” BloombergNews, Jan. 6, 2012.

Figure 1. Some summary statistics on the annual growth of factors contributing to data explosion.

Annual growth statistics

Mobile video Total traffic (2011) 62% 34%Streaming CAGR (2011 - 16)

Source: Cisco VNI 2011 - 2016 (http://www.tinyurl.com/VNI2012)

Cloud services Consumer traffic (2011) 83% 67%Consumer traffic CAGR (2010 - 15)

Source: Cisco global cloud index 2010 - 15 (http://www.tinyurl.com/CiscoCloud2010)

Data-hungry apps Facebook (2010) 267% 87%Skype (2010)

Source: Allot global mobile trends report 2011 (http://money.cnn.com/2011/02/08/technology/smartphone_data_usage/index.htm)

High-resolution devices iPad2 1024 x 768 pixels 2048 x 1536 pixelsiPad LTE

Source: Apple (http://www.apple.com/ipad/compare/)

Growth in mobile data demand

Mobilevideo

Cloudservices

Data-hungryapps

High-resolutiondevices

Rapid growth indemand for mobile data

The widespread

adoption of

handheld devices,

equipped with

powerful processors,

high-resolution

cameras, and larger

displays, has made it

convenient for

users to stream

high-quality videos

and exchange large

volumes of data.

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

______

IEEE Communications Magazine • November 2012 93

significant factor in traffic growth for ISPs andexpensive monthly bills for consumers.9 Similar-ly, machine-to-machine (M2M) applications thatgenerate data intermittently (e.g., sensors andactuators, smart meters) or continuously (e.g.,video surveillance) often load the network withlarge signaling overhead [2]. Yet the intrinsictime elasticity of such services also creates anopportunity for shifting them to low-congestiontimes by applying appropriate price incentives.

CAPACITY-HUNGRY APPLICATIONSThe popularity of handheld devices has also ledto rapid growth in the development of band-width-hungry applications for social networking,file downloads, music and video streaming, per-sonalized online magazines, and so on. Forexample, Allot Communications reported that in2010, traffic from services like Skype grew about87 percent, and Facebook by 267 percent.10 Vir-gin Media Business reports that the averagesmartphone software uses 10.7 Mbytes/h, withthe highest-usage app, Tap Zoo, consuming upto 115 Mbytes/h.11

These developments have led ISPs to viewpricing as their ultimate congestion managementtool, paving the way for the adoption of SDPmeasures, such as time-dependent and usage-based pricing.

TRENDS IN PRICING FORCONGESTION MANAGEMENT

FLAT-RATE PRICINGAOL’s historic switch12 in 1996 from hourlyrates to an unlimited flat rate data plan madethe latter the dominant model for access pricing.However, as shown in Fig. 2, there has been adrastic shift in ISPs’ pricing policies since 2008in both wireline and wireless networks.13 Com-cast’s data cap of 250 Gbytes was followed bysimilar measures for AT&T’s U-Verse and digi-tal subscriber line (DSL) services in 2011. Unlim-ited data plans for mobile services came to thesame end: between 2010 and 2012, AT&T, Veri-zon, and T-Mobile all eliminated their unlimitedplans in favor of tiered data plans. While AT&Tand Verizon embraced usage-based overagesabove data caps, T-Mobile adopted bandwidththrottling for users who exceeded their monthlycap.

USAGE-BASED PRICINGSince 2008, the slow demise of flat-rate plans hasbeen widely covered in the media (Fig. 2). InAugust 2008, Comcast implemented a 250 Gbytecap on its cable network per household, threat-ening repeat offenders (i.e., those who exceedthe caps twice within six months) with termina-tion for one year [3, 8]. Time Warner conductedUBP trials in Beaumont, Texas, which offered itscustomers a 768 kb/s line with 5 Gbyte cap for$29.95/mo or a 15 Mb/s line with 40 Gbyte capfor $54.90/mo, both with $1/Gbyte overage fees.Although these met with poor reception, a simi-lar UBP trial by AT&T from November 2008 toApril 2010, conducted in Beaumont and Reno,Nevada, fared better [3]. By May 2011, AT&Tintroduced a monthly data cap of 250 Gbytes on

U-Verse and 150 Gbytes for DSL lines with $10overage for additional blocks of 50 Gbytes.

For wireless networks, the transition to UBPhas been faster. In 2010, AT&T introducedthree tiered data plans of $15 fees for 200Mbytes, $25 for 2 Gbytes, and $45 for 4 Gbytes;exceeding the caps resulted in $10/Gbyte over-age fees for the 2 and 4 Gbyte plans, and $15 foreach additional 200 Mbytes for the lowest tierplan [3]. Within a year, AT&T also started throt-tling the heaviest 5 percent of its users, eventhose with unlimited data plans [8]. As of July2012, all major ISPs, including Comcast, AT&T,Verizon, and T-Mobile, have moved to someform of UBP; even Sprint, the only large carrierwith an unlimited data plan, cancelled it for lap-tops and notebooks.14 Next, we discuss some ofthe main criticisms and implications of replacingflat rate plans by UBP [3, 7].

APPLICATION-BASED PRICINGAlthough highly contentious from a networkneutrality standpoint, as discussed next, app-based pricing is emerging as a new trend in someEuropean and Asian markets [8]. App-basedpricing can come in several forms, such as zero-rating (toll-free) or sponsored content. Underthese plans, either the ISP or content providerpays for access to certain popular applications(e.g., Mobistar, a Belgian ISP, has a zero-rateplan for Facebook, Twitter, and Netlog). Anoth-er form of app-based pricing is content andaccess bundling; for example, the Danish opera-tor TDC bundled its music streaming service,TDC Play, into its data plans [8]. In the UnitedStates, cable provider Comcast recently cameunder scrutiny for not counting its own Xfinityfor Xbox video-on-demand service toward users’data caps.15 We discuss the technological andregulatory perspectives of app-based bundling ingreater detail later.

TIME-DEPENDENT PRICINGIn the telecom industry, time-dependent pric-

ing has long been practiced by ISPs, often assimple two-period (peak and off-peak) pricingfor voice calls. For example, both AT&T andVerizon, the two largest wireless operators in theUnited States, offer free calling on nights andweekends, while capping minutes used duringthe day. But today, operators in price-sensitiveregions such as India and Africa have begun tooffer more innovative dynamic TDP for voicecalls. MTN Uganda, for instance, has imple-mented a dynamic tariffing plan16 in which theprice to make a voice call changes hourlydepending on the congestion conditions at thecall’s originating location. While these TDPinnovations have been applied only to voice traf-fic, there has been a system implementation andpilot trial of TDP for mobile data traffic [9, 10].

REGULATORY AND RESEARCHPERSPECTIVES ON PRICING TRENDS

As more innovative pricing practices are pro-posed by researchers and adopted by ISPs, theyshould be analyzed with regard to their efficacyin meeting today’s challenges, as well as their

9 D. Ngo, “iCloud: TheHidden Cost for theMagic, and How to AvoidIt,” Cnet News, Nov. 7,2011.

10 D. Goldman, “You areUsing More Data ThanYou Think,” CNNMoney, Feb. 8, 2011.

11 K. C. Tofel, “DataHungry Mobile Apps Eat-ing into Bandwidth Use,”Gigaom, May 19, 2011.

12 M. Grebb, “WashingtonState calls AOL on Flat-Rate Plan,” Wired, Nov.22, 1996.

13 References to newsitems discussed in thissection can be found inFig. 2.

14 J. O. Gilbert, “SprintCancels Unlimited Datafor Hotspots, Tablets andNetbooks,” HuffingtonPost, Oct. 21, 2011.

15 D. Mitchell. “Is Com-cast Violating Net-Neu-trality Rules?” CNNMoney, May 16, 2012.

16 T. Standage, “TheMother of Invention,”Special Report on MobileMarvels, The Economist,Sept. 24, 2009.

SDP is in fact a

broad set of smart

mechanisms and

ideas, which includes

fair throttling, smart

demand shaping,

intelligent WiFi

offloading, quota-

aware content distri-

bution, sponsored

access, and so on.

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

IEEE Communications Magazine • November 201294

implications on a free and open Internet. Thissection discusses academic research that seeks toaddress some of the key questions and concernsrelated to pricing policies. An overview of thekey viewpoints on these different pricing plans isprovided in Table 1.

IS USAGE-BASED PRICING USEFUL?ISPs like Comcast, AT&T, and Verizon haveargued to the FCC that UBP is necessary forinvesting in network infrastructure, so as tomatch their revenues to the cost incurred fromthe explosive growth in data usage [6]. The opti-mal UBP mechanism that maximizes ISP rev-enue in a resource-constrained network withincomplete user information was presented in

[11]. While UBP can be beneficial as a conges-tion control mechanism when growth in demandexceeds the rate at which extra capacity can beadded, its effectiveness depends on its imple-mentation [3]. Odlyzko et al. argue in [3, 12] thatdue to the statistical multiplexing principle ofthe Internet, bandwidth should not be viewed asa “consumable” good that is unfairly appropriat-ed by bandwidth hogs. Moreover, they arguethat the fiber deployment initiatives of AT&Tand Verizon were motivated more by competi-tion with cable operators’ triple-play servicesthan by any real threat of congestion. In the caseof wireless networks, where congestion at basestations is indeed an issue,17 the effectiveness ofUBP is also under question [3, 7]. The key criti-

Figure 2. Key events in the evolution of ISP pricing practices between 2008–2012.

Wireline

1. R. Paul, “40GB for $55 per month: Time Warner bandwidth caps arrive,” ArsTechnica, June 8, 2008.2. Y. Adegoke, “Comcast to limit customer’s broadband usage”, Reuters, August 28, 2008.3. B. Reed, “Bandwidth caps coming to AT&T wireline services,” Network World, March 14, 2011.4. A.Lee, “AT&T to begin broadband caps,” Huffington Post, May 2, 2011.5. D. Goldman, “Comcast scraps broadband cap, moves to usage-based billing,” CNN Money, May 17, 2012.

Wireless

6. J. Cox, “AT&T shifts to usage-based wireless data plans,” Network World, June 2, 2010.7. B. Reed, “T-Mobile’s data cap embrace leaves Sprint as lone ‘unlimited’ 4G carrier,” Network World, May 23, 2011.8. L. Segall, “Verizon ends unlimited data plan,” CNN Money, July 6, 2011.9. S. Lawson, “AT&T to throttle big users of unlimited data,” Computer World, July 29, 2011.10. K.C. Tofel, “Apple’s iPad 4G LTE plan pricing detailed: Watch your speeds!” GigaOm, March 7, 2012.11. J. Gilbert, “Verizon unlimited data plans for grandfathered customers to end soon,” Huffington Post, May 16, 2012.12. R. Cheng, “Why Verizon’s shared data plan is a raw deal,” CNet News, June 12, 2012.

Wireless

Wireline

Introduction ofusage-based penalties

Elimination of unlimited data plans

Introduction of data caps

2008 2009 2010 2011 2012

Verizon eliminates newunlimited smartphone plans

(July 2011)8

AT&T caps U-verse to 250Gbytes and DSL to 150

Gbytes with a $10 penaltyfor each additional 50 Gbytes

per month (May 2011)4

AT&T starts throttlingunlimited iPhone users

(July 2011)9

AT&T starts charging a$10/Gbyte overage fee

for smartphone dataplans (June 2010)6

T-Mobile startsdata caps and

throttling penalty(May 2011)7

Time-Warnertrials data

caps in Texas(June 2008)1

Comcast capsdata at 250 Gbytes

(August 2008)2

Comcast movestoward tiered usage-

based billing(May 2012)5

AT&T startsthrotting wireline

users(April 2011)3

No unlimited plansoffered for iPad LTE

(March 2012)10

Verizon introducesshared data plans

with unlimitedvoice and text(June 2012)12

Verizon to eliminateunlimited data plans

(May 2012)11

17 G. Fleishman, “TheState of 4G: It’s All AboutCongestion, Not Speed,”Ars Technica, Mar. 29,2010.

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

IEEE Communications Magazine • November 2012 95

cism is that UBP cannot alleviate peak conges-tion unless it contains a time-dependent pricingcomponent to provide dynamic feedback on thenetwork congestion level [10].

DOES APP-BASED PRICINGTHREATEN NET-NEUTRALITY?

Reducing congestion requires investment in thenetwork infrastructure, but questions remain onwhom to charge to subsidize this expense. Giventhat the consumers, even “bandwidth hogs,”impose a relatively small marginal operationalcost on the network, Odlyzko et al. [3] haveargued that pricing mechanisms like UBP canoften be misused as a tool to increase return oninvested capital. On the other hand, formerAT&T CEO Ed Whitacre generated a new debateon net-neutrality by suggesting that in addition toconnectivity fees, ISPs should be allowed tocharge content providers a part of ad revenues inexchange for the right to access end users.18

Recent research by Musacchio et al. [4] hasshown that two-sided pricing (i.e., a “non-neutral”network) can sometimes increase social welfare,particularly when the ratio of certain modelparameters characterizing advertising rates andend-user price sensitivity is either high or low. Incontrast to this view, Economides [5] has arguedthat such ability would lead to paid prioritization

arrangements in which app and content providerscould pay ISPs to prioritize their packets overcompetitors’. This would create market inefficien-cies and hurt innovation, and can even incentivizeISPs to inflate network congestion problems inorder to extract higher revenues from content pro-viders [5]. Similarly, agreements between ISPs andcontent providers on bundling applications andaccess or zero-rating specific applications areviewed as non-neutral, discriminatory practices [3].

CAN TIME-DEPENDENT PRICINGHURT DEMAND?

One of the main arguments in favor of flat ratepricing as opposed to UBP or TDP has beenthat consumers are willing to pay more for flat-rate plans. Flat-rate plans encourage higher par-ticipation (i.e., a positive network externality),which benefits the Internet ecosystem [3]. Vin-ton Cerf suggests that time-dependent variants,like UBP with off-peak discounts, can also “endup creating the wrong incentives for consumers toscale back their use of Internet applications overbroadband networks” [7]. But in contrast to theviews expressed above, dynamic TDP for voicecalls by MTN Uganda actually increased thedemand of their price-sensitive demographic.14

Broadband analytics firm Wireless Intelligencereports that every one U.S. cent decrease in the

18 E. Whitacre, “At SBC,It’s All About “Scale andScope,” edited by P.O’Connell, BusinessWeek Mag., Nov. 6,2005.

Table 1. Summary of features and opinions on some current pricing policies.

Pricing Policy Arguments in Favor Arguments Against Adoption Trends Neutrality Perspectives

Flat-rate(unlimited)

• Simpler, preferred by con-sumers• Users are typically more will-ing to pay higher flat rate fees.

• Unsustainable because· of“bandwidth hogs.”• Light users subsidize heavyusers.• Encourages wastage.

• Traditional model• Mostly discarded by USwired and wireless ISPs likeComcast, AT&T, Verizon, T-Mobile.

• Modeling flat-rates as a form ofbundling -shows the benefit of het-erogeneous user‘s preferences,despite disproportionate usage vol-umes [12].• Flat pricing can cause loss of usernet utility, particularly when the con-sumers have low price sensitivity [16].

Usage-based(UBP: data caps,throttling, over-ages)

• Allows more efficient use ofbandwidth.• Can lower cost for consumersas they pay in proportion totheir consumption.• Helps ISPs to match their rev-enues to costs.

• Lacks the temporal dimen-sion needed to solve ISPs’ peakcongestion issue.• Can reduce consumer usageand spending.

• In 2008, Comcast intro-duced a data cap of250GB/month per house-hold.• T-Mobile and AT&T throt-tle heaviest users.• Verizon and AT&T botheliminated flat-rate in lieu oftiered data plans with$10/GB overages.

• Data caps are less efficient thantransmission caps [7].• UBP charges users regardless ofnetwork congestion at any giventime [3, 7].• Network maintenance costs are notdirectly relatable to traffic volume [3].• ISPs’ revenue losses due to regu-lations against price discriminationcan be mitigated by offering pricesnonlinear in data volume or datarate, with discounts on higherdata-rate demand [16].

App-based(zero-rating,toll-free apps)

• Offers consumers choices forpersonalized plans.• App bundling helps ISPs togain consumers.• Content-provider subsidies(toll-free) benefit consumers.

• Discriminatory, non-neutral• Favors big content and appproviders.• Deep Packet Inpsection forapp classification has privacyimplications.

• Bundled subscriptions inEurope by Tele-DanmarkCommunications.• Zero-rated access to Face-book, Twitter, Netlog byMobistar in Belgium.• Access and app bundlingby Three in the UK.

• Last-mile termination feesimposed on third party content orapplication providers will lead topaid prioritization over competingservices, harming innovation andcreating market inefficiencies [5].• Inability to price discriminateacross flows causes higher revenuelosses when users are less pricesensitive [16].

Time-dependent(TDP: static,dynamic)

• Exploits time-elasticity ofdemand by creating incentivesfor users to shift demand.• Can reduce peak demand andfill up valley periods.

• Can penalize light, low-income users who cannot shifttheir demand.• Needs improved user inter-face for price notifications.

• Exists as daytime/night-time discount.• Already used in electricity,transport markets.• Dynamic TDP for voicecalls is practiced in India,Africa.

• Empowers users with the choiceof scheduling elastic traffic to saveon monthly bills instead of flat-rateor volume caps [9, 10].• Impact of TDP on applicationecosystem needs further studiesand trials [10].

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

IEEE Communications Magazine • November 201296

effective price per minute of voice calls leads toa 6.9 and 13.5 times increase in minutes usedper month for developed and developing coun-tries, respectively.19 With respect to mobiledata, initial TDP trials (as discussed earlier)have found that TDP increases the overalldemand, partly due to higher usage in discount-ed periods [9, 10].

INCENTIVE MODELS FORTIME-SHIFTING OF DATA

Shifting demand from congested periods to off-peak times requires pricing policies with the rightincentives to induce such a response from users.20

Some of these policies (e.g., peak-load pricing,off-peak discounts, day-ahead pricing, and real-time pricing) have an explicit time-dependentpricing feature; others, like game-theoretic (auc-tion) models and prioritization-based models(e.g., smart market, raffle-based, token bucketpricing), use more implicit time-varying incentivesto induce the time-shifting of data to less congest-

ed periods. An overview of the features, contribu-tions, and limitations of the incentive models dis-cussed below is provided in Table 2.

PEAK LOAD PRICINGParris et al. [13] considered a time-dependentpricing model for peak load pricing (PLP) inwhich specified periods of time are pre-classi-fied as peak and off-peak hours. In their simu-lation-based study, the authors consider ahigher arrival rate and a higher fee during thepeak periods. Each user request has an associ-ated elasticity that determines whether arequest arriving during a peak period can bedeferred to a subsequent off-peak period. Theauthors show that PLP can spread out demandover time periods and generate higher revenuesthan non-peak load pricing, but it can also seg-ment the user base, with low-budget and low-elasticity users being denied service. Thispotential adverse impact of PLP is similar tothat of UBP (as observed by Odlyzko et al. in[3]) in that it hurts broadband adoption in low-income groups.

Table 2. Comparison of features, contributions, and limitations in representative research works.

Incentive Schemes Key Model Features Contributions Simulation/Systems Disadvantages

Peak load pricing(PLP)(Parris et al. [13])

• Higher prices and higherarrival rate of user requestsin pre-classified peak hours.• Elasticity of user requestsdetermines if the demand isdeferred to off-peak hours.

PLP can flatten out dailydemand and generateshigher revenue than non-PLP, while also reducingcall blocking probability.

Simulations are used toverify the effectiveness ofPLP in reducing peak uti-lization in an integratednetwork.

Authors show that PLPcan segment the userbase, with low-budgetand low-elasticity usersbeing denied service.

Off-peak discount(EI-Sayed et al. [2])

Studies the impact of off-peak window size, discounts,and usage shifted on thecost savings and revenues.

Provides several simulationresults on capacity savingsfrom peak load shifting.

The observations arebased on numericalinvestigations for arange of scenarios.

Does not include a realworld deployment ortrial results to validate orquantify the benefits.

Dynamic day-aheadpricing(Ha et al. [9, 10])

Optimized “day-ahead” TDPfor mobile data computed inan ISP-user feedback controlloop.

In trial results, the maxi-mum peak to average ratioof usage volume decreasesby 30%.

Conducts the first systemimplementation and trialof TDP for mobile data.

Requires careful designof user interfaces anddata usage notificationsystem.

Real time pricing(MacKie-Mason andVarian [14])

• Dynamic adaptation ofprices to congestion level.• ISP admits packets basedon user bids for auction.

Introduces “Smart market”for real-time congestionpricing.

Provides analyticalresults, but does notinclude a real trial of thesystem.

Requires automation forend-users to bid onpackets during congest-ed times.

Game-theoreticanalysis(Jiang et al. [17])

(Shen and Basar[15])

• Introduces a parameterizedmodel of users choosingtheir usage time based oncongestion, price offers, andintrinsic preferences.

• Uses non-cooperativegames to study user incen-tives from a mechanismdesign perspective.

Studies Price of Anarchy,i.e., social cost of ISP offer-ing profit-maximizingprices instead of maximiz-ing social welfare.

Shows that ISP profits canbe larger with non-linearusage fees.

Numerical examplesshow cases when ISPs’optimal prices may beextremely socially subop-timal.

Numerical results showISP profits increase morethan 38% with nonlinearpricing over linear fees.

Considers only a singleservice provider scenario.

Marketing of non-linearpricing structure to con-sumers can be difficult.

Token bucket pric-ing(Lee et al. [18])

Users receive a fixed numberof tokens that can be used toobtain higher QoS duringcongested times.

Shows that token pricinggives higher utility to usersthan flat rate for both sin-gle and two service classes.

Numerical studies showthat token bucket pricingincreases social welfarerelative to flat rate.

Requires automation tohelp users follow theoptimal token usage pol-icy.

Raffle-based pricing(Loiseau et al. [19])

Lottery-based rewards inproportion to users’ contri-bution to peak reduction.

Shows that the equilibriumprices maximize the totalsocial welfare.

Numerical studies showrobustness of social wel-fare to changes in userwillingness to offload.

ISPs need to measurehow much each useroffloads, and user com-pensation depends onlottery.

19 http://www.wirelessin-telligence.com/analy-sis/2011/09/how-pricing-dynamics-affect-mobile-usage/

20 Economic incentivesgovern the actions of usersand ISPs, and forms acore of the “economiclayer” of the network asdescribed by J. Walrand in“Economic Models ofCommunication Net-works,” Ch. 3, Perfor-mance Modeling andEngineering, Ed. Z. Liu,Springer, 2008.

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

____________

__________________________

________________

IEEE Communications Magazine • November 2012 97

OFF-PEAK PRICE DISCOUNTS

Off-peak discounting from a predeterminedbaseline price is a logical dual of peak-load pric-ing. El-Sayed et al. [2] numerically studied time-dependent price discounts for mobile data. Theirmodel includes three parameters to quantify theimpact of TDP on both network cost savings andrevenue loss:• The duration of the off-peak window in

which off-peak incentives are offered• A discount percentage per megabyte• The percentage of busy hour load that the

users shift due to off-peak incentivesThe basic idea of offering TDP off-peak dis-counts is similar to later works on dynamic TDP[9, 10], though these latter works also include ananalytical model of a control-feedback loop thatadjusts the offered discounts and presents a real-world trial.

REAL-TIME PRICINGThe time-shifting of demand can also be real-ized by providing end users with an explicitpricing signal about the real-time network con-dition. MacKie-Mason and Varian [14] presentthe idea of a “Smart Market,” which is a closedcontrol-feedback loop in which the networkadapts the prices to congestion levels. In thisapproach, each user places a “bid” on eachpacket that reflects her willingness to pay tosend the packet onto the network at a giventime. The gateway admits packets in descendingorder of their bids as long as the network per-formance remains above a desired threshold.Users are charged according to the minimumbid on a packet admitted into the network atthe time, and thus pay only for the congestioncost at the market-clearing price. Auction-based mechanisms help users to explicitly speci-fy their willingness to pay for handling packetsfrom applications with different time elastici-ties. A limitation of this approach is that itrequires intelligent automation on the clientside to relieve the end users’ burden of decisionmaking.

GAME-THEORETIC PRICING

Game theory offers a natural, but somewhat styl-ized, model for time-dependent pricing, as itallows one to consider consumers’ utility in theamount of usage shifted in response to offeredprices. Shen and Basar [15] use non-cooperativegames to study user incentives from the perspec-tive of mechanism design. Although their workdoes not consider time-dependent pricing, theyshow that while most service providers use linear(i.e., usage-based) charging, nonlinear chargingcan yield large increases in ISP profit. Theirresults are relevant in the context of observa-tions made by Hande et al. [16] that nonlinearpricing can also reduce ISPs’ revenue losses,even in regulatory environments that prohibitexplicit price discrimination across consumers.

Jiang et al. [17] study hourly time-dependentpricing offered by a selfish ISP, comparing theprofit-maximizing time-dependent prices to thesocially optimal ones. To compute these prices,the authors assume that users choose the timingof their Internet usage based on the congestioncondition, price offered, and intrinsic preferencefor that time slot. They then focus on the priceof anarchy, defined in terms of the social welfareunder profit-maximizing time-dependent priceswhen compared to the maximum social welfare.If several low-utility users are present in the net-work and the ISP cannot offer different prices tothese users, the price of anarchy may be arbitrar-ily large (i.e., the ISP-optimal prices may beextremely socially suboptimal). This result sug-gests that some regulatory price restrictions canimprove social welfare.

TOKEN BUCKET PRICINGToken bucket pricing [18] divides the day intopeak and off-peak hours. Users can accumulatedaily tokens, which may be exchanged for serviceduring peak hours; without sufficient tokens,users must wait until off-peak hours to consumedata. Viewing the tokens as currency, one seesan analogy with time-dependent pricing [2, 9,14]; the principal difference is that users’ bud-

Figure 3. TUBE Trial participants a) could view future prices (color-coded), along with price and usage his-tory. They responded by b) increasing their usage more in low-price, relative to high-price periods.

Weighted average change in low price periods (%)0-100

0

-100

Wei

ghte

d av

erag

e ch

ange

in h

igh

pric

e pe

riod

s (%

)100

200

300

400

500

600

-50-100

-50

-100

0

50

0 50

100 200 300

(a) (b)

400 500 600 700 800

Auction-based mech-

anisms help users to

explicitly specify their

willingness to pay for

handling packets

from applications

with different time

elasticities. A limita-

tion of this approach

is that it requires

intelligent automa-

tion on the client

side to relieve the

end users’ burden of

decision making.

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

IEEE Communications Magazine • November 201298

gets with token pricing are determined by thefixed rate at which tokens are received, insteadof the variable amounts of money they are will-ing to spend on peak-hour usage. The presenceof these token-based incentives is shown toincrease social welfare relative to flat-rate unlim-ited data plans. However, token pricing mayrequire client-side automation for users to follow“optimal” policies for using their tokens. More-over, token-based pricing may require severaldifferent plans, corresponding to different ratesof token accumulation. For instance, a businessuser may be willing to pay a higher flat fee inexchange for more tokens that can be used dur-ing congested hours of the day.

RAFFLE-BASED PRICINGRaffle-based pricing also divides the day intotwo time periods: peak and off-peak. Users arethen offered probabilistic incentives to shift theirdemand to off-peak periods, in the form of a raf-fle or lottery for a given monetary reward. Theprobability of winning the lottery is proportionalto the user’s contribution to the reduction inpeak demand. This “all-or-nothing” lottery mayinstead be replaced by one in which the ISPdivides the total reward by the total amount oftraffic shifted, paying this amount to each userwith a probability equal to the percentage ofusage shifted to the off-peak period. In [19],Loiseau et al. derive expressions for the Nashequilibrium of this user-ISP interaction andshow that in some cases the social welfare ismore robust to price variations than a compara-ble time-of-day pricing plan with two periods.Yet the uncertainty inherent in raffle-based pric-ing may reduce its effectiveness; users may notshift their usage to off-peak periods withoutguaranteed rewards. This uncertainty is exacer-bated by the reward’s dependence on otherusers’ behavior; hence, a field trial of such a pol-icy would be necessary to understand its efficacy.

DYNAMIC DAY-AHEAD PRICINGHa et al. [10] introduce Time-Dependent Usage-Based Broadband Price Engineering (TUBE;http://www.datami.com), a fully integrated TDPsystem that uses “day-ahead pricing,” a schemeoften used in electricity markets.21 In this dynam-ic TDP policy, the prices vary by the hour, butusers are informed of the prices one day inadvance. This advance notice alleviates the useruncertainty inherent in real-time pricing, allow-ing users to plan their bandwidth consumptionover the next day if desired. TUBE models thisuser behavior as shifting mobile data usage fromone period to another in response to futureprices. Users have a certain probability of shift-ing their traffic from each period to each otherperiod, as a function of the time differencebetween the periods and the price in the laterperiod. Given functions describing these proba-bilities, the ISP can calculate the traffic patternover the day as a function of the prices offeredand use this calculation to optimize its profit.These prices are offered to users on their indi-vidual devices; the ISP then records the resultingusage and revises its estimates of users’ respons-es to offered prices.

TUBE’s principal advantage over PLP [13]

and off-peak discounts [2] is its dynamic adjust-ment of future prices in response to changes inusage behavior; TUBE thus performs better inreducing peak congestion [10] than does staticTDP. Additionally, to the best of our knowledge,TUBE is the first functional TDP system formobile data that has undergone a real customertrial22 [9, 10]. The TUBE project has alsoreleased a free personalized bandwidth manage-ment app, called DataWiz, for iOS and Androiddevices for the general public (http://scenic.princeton.edu/datawiz/).

FEASIBILITY STUDIESAs discussed in the previous sections, incentiviz-ing time-shifting of data has been practiced inthe form of static peak and off-peak discountsfor voice calls. But more dynamic forms of TDPare emerging in growing economies; MTN Ugan-da’s dynamic tariffing plan for voice calls result-ed in a volume increase of 20–30 percent,indicating that price incentives for voice calls canstrongly influence user behavior.

The potential benefits of TDP can be higher inthe case of mobile data, since many applicationstoday (e.g., online backups in the cloud, app andmovie downloads, peer-to-peer) have an intrinsictime elasticity that can be effectively exploited[10].23 TUBE successfully induced consumers toshift their data demand to lower-priced, less con-gested periods [9, 10]. This nine-month trial con-sisted of 50 participants with iPads or iPhones andAT&T 3G subscriptions. Trial participants werecharged for mobile 3G data usage according toTUBE’s day-ahead pricing algorithms. As shown inFig. 3a, participants could view their price andusage history, as well as future prices, on their iOSdevices. In one phase of the trial, the hourly pricesalternated between high and low, to test users’price sensitivity. Figure 3b shows the resultingchange in usage, measured relative to usage withstatic UBP of $10/Gbyte. The reference line showsan equal percentage change in usage for low- andhigh-price periods, with each data point corre-sponding to the weighted average of one user’spercentage changes in usage in all high-price andall low-price periods. Since most data points liebelow the reference line, most users increasedtheir usage in low-price relative to high-price peri-ods, indicating some price sensitivity. When opti-mized time-dependent prices were offered, thisprice sensitivity helped bring down the peakdemand, measured in terms of the peak-to averageratio (PAR) over each day. The maximum PARdecreased by 30 percent, while the overall usageincreased by 130 percent due to higher demand inthe valley periods with large discounts [10].

CONCLUSIONSThe growing fear of a data tsunami, particularlyin wireless networks, has led ISPs to use pricingas a congestion management tool. However,understanding the potential benefits and risksassociated with different pricing schemes is cru-cial in both choosing effective data plans andguiding regulatory decisions in maintaining anopen Internet. For more than two decades,research in network economics has shown that

21 CNET Energy Pricing.http://www.cntenergy.org/pricing/

22 The Berkeley INDEXproject of the 1990s (P. P.Varaiya, R. J. Edell, andH. Chand, INDEX Pro-ject Proposal, 1996)attempted a similar initia-tive for wired networkpricing.

23 Dynamic day-aheadpricing has also been suc-cessfully practiced forsome years in the electrici-ty industry, e.g., OntarioIESO(http://www.ieso.ca/imoweb/marketData/market-Data.asp). Viewing differ-ent mobile applications asanalogous to applianceswith different delay toler-ances, one sees a clearanalogy between thedemand-response modelsof time-dependent pricingfor mobile data and elec-tricity market.

TUBE is the first

functional TDP

system for mobile

data that has

undergone a real

customer trial.

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

_____________

_____

_____________

_____

___

IEEE Communications Magazine • November 2012 99

providing the right incentives for consumers totime-shift their demand, as commonly done withtime-dependent pricing, is the key to resolvingthe issue of network congestion. In this work, weprovide an overview of various incentive mecha-nisms for the time-shifting of demand, includingsome results from very recent works and con-sumer trials, with the aim of helping researchersand practitioners become aware of the variouschallenges and opportunities arising from today’snetwork congestion problems. The evolution ofnew technologies like 4G/LTE, smart grids, andcloud computing will greatly depend on the pric-ing policy choices that we make today.

ACKNOWLEDGEMENTSWe would like to specially thank AndrewOdlyzko, Roch Guerin, Nicholas Economides,Rob Calderbank, Krishan Sabnani, ShyamParekh, Sundeep Rangan, Victor Glass,Prashanth Hande, Raj Savoor, Steve Sposato,Patrick Loiseau, and all the participants of the2012 Smart Data Pricing (SDP) workshop inPrinceton, for sharing their valuable commentsand insights. We also acknowledge the supportof our industrial collaborators, in particular,Reliance Communications, MTA Alaska, AT&T,Qualcomm, Comcast, Mediacom, and theNational Exchange Carrier Association.

REFERENCES[1] Cisco Visual Networking Index, White Paper, Feb. 2012,

available online at http://www.cisco.com/ en/US/solu-tions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11-520862.pdf, 2012.

[2] El-Sayed et al., “Mobile Data Explosion: Monetizing theOpportunity through Dynamic Policies and QoS Pipes,”Bell Labs Tech. J., vol. 16, no. 2, 2011, pp. 79–100.

[3] A. Odlyzko et al., “Know Your Limits: Considering theRole of Data Caps and Usage Based Billing in InternetAccess Service,” Public Knowledge White Paper,http://publicknowledge.org/files/UBP%20paper%20FINAL.pdf, 2012.

[4] J. Musacchio, G. Schwartz, and J. Walrand, “A Two-Sided Market Analysis of Provider Investment IncentivesWith an Application to the Net-Neutrality Issue,” Rev.Network Economics, vol. 8, no. 1, Berkeley ElectronicPress, 2009.

[5] N. Economides, “ Why Imposing New Tolls on Third-Party Content and Applications Threatens Innovationand Will Not Improve Broadband Providers’ Invest-ment,” NET Institute Working Paper No. 10-01; NYULaw and Economics Research Paper No. 10–32, 2010.

[6] S. Higginbotham, “FCC Opens the Door forMetered Broadband,” Dec. 1, 2010, http://gigaom.com/2010/12/01/fcc-opens-the-door-for-metered-web-access/, 2010.

[7] V. Cerf, “What’s A Reasonable Approach for ManagingBroadband Networks?,” Google Public Policy Blog,August 4, 2008, http://googlepublicpolicy.blogspot.com/2008/08/whats-reasonable-approach-for-manag-ing.html, 2008.

[8] S. Sen et al., Pricing Data: Past Proposals, Current Plans,and Future Trends, http://arxiv.org/pdf/1201.4197.pdf,2012.

[9] S. Ha et al., “Demo: Enabling Mobile Time-DependentPricing,” Proc. ACM Mobisys, Lake District, U.K., 2012.

[10] S. Ha et al., “TUBE: Time Dependent Pricing for MobileData,” Proc. ACM SIGCOMM, Helsinki, Finland, 2012.

[11] S. Li, J. Huang, and S.-Y. R. Li, “Revenue Maximizationfor Communication Networks with Usage-Based Pric-ing,” Proc. Globecom, Honululu, HI, 2009.

[12] P. Nabipay, A. Odlyzko, and Z.-L. Zhang, “Flat VersusMetered Rates, Bundling,” and “Bandwidth Hogs,”Proc. NetEcon, San Jose, CA, 2011.

[13] C. Parris, S. Keshav, and D. Ferrari, A Framework forthe Study of Pricing in Integrated Networks,Tech. Rep.,Tenet Group, ISCI, UC Berkeley, 1992.

[14] J. MacKie-Maso and H. Varian, Pricing the Internet,Public Access to the Internet, B. Kahin and J. Keller,Eds., Prentice-Hall, 1995.

[15] H. Shen and T. Basar, “Optimal Nonlinear Pricing for aMonopolistic Network Service Provider with Completeand Incomplete Information,” IEEE JSAC, vol. 25, no. 6,2007, pp. 1216–23.

[16] P. Hande et al., “Pricing under Constraints in AccessNetworks: Revenue Maximization and Congestion Man-agement,” Proc. IEEE INFOCOM, San Diego, CA, 2010.

[17] L. Jiang, S. Parekh, and J. Walrand, “Time-DependentNetwork Pricing and Bandwidth Trading,” Proc. IEEENOMS Wksp., Salvador, Brazil, 2008.

[18] D. Lee et al., “A Token Pricing Scheme for InternetServices,” Proc. 7th ICQT, Paris, France, 2011.

[19] P. Loiseau et al., “Incentive Schemes for Internet Con-gestion Management: Raffle versus Time-of-Day Pric-ing,” 2011 Annual Allerton Conf., Urbana-Champaign,IL, 2011.

BIOGRAPHIESSOUMYA SEN [S’08, M’11] received a B.E. (Hons.) in electron-ics and instrumentation engineering from BITS-Pilani, India,in 2005, and both M.S. and Ph.D. degrees in electrical andsystems engineering from the University of Pennsylvania in2008 and 2011, respectively. He is currently a postdoctoralresearch associate at Princeton University. He has beenactively involved in forging various industrial and academicpartnerships, and serves as a reviewer for several ACM andIEEE journals and conferences. He is also a co-organizer ofthe Smart Data Pricing (SDP) Forum 2012 and GLOBECOM2012 SDP Industry Forum, which promotes industry-aca-demic interaction on pricing policy research. His researchinterests are in Internet economics, communication sys-tems, social networks, and network security. He has pub-lished several research articles in these areas and won theBest Paper Award at IEEE INFOCOM 2012.

CARLEE JOE-WONG [S’11] is a graduate student at PrincetonUniversity’s Program in Applied and Computational Mathe-matics. She received her A.B. (magna cum laude) in mathe-matics from Princeton University in 2011. Her researchinterests include network economics and optimal controltheory, and her work on the fairness of multi-resource allo-cations received the Best Paper Award at IEEE INFOCOM2012. In 2011, she received the NSF Graduate Research Fel-lowship and the National Defense Science and EngineeringGraduate Fellowship.

SANGTAE HA [S’07, M’09, SM’12] is an associate researchscholar in the Department of Electrical Engineering atPrinceton University, leading the establishment of thePrinceton EDGE Laboratory as its associate director. Hereceived his Ph.D. in computer science from North CarolinaState University in 2009 and has been an active contributorto the Linux kernel. During his Ph.D. years, he participatedin inventing CUBIC, a new TCP congestion control algo-rithm. Since 2006, CUBIC has been the default TCP algo-rithm for Linux and is currently being used by more than40 percent of Internet servers around the world and byseveral tens of million Linux users for daily Internet com-munication. His research interests include pricing, greening,cloud storage, congestion control, peer-to-peer network-ing, and wireless networks.

MUNG CHIANG [S’00, M’03, SM’08, F’12] is a professor ofelectrical engineering at Princeton University, and an affili-ated faculty member in applied and computational mathe-matics, and computer science. He received his B.S. (Hons.),M.S., and Ph.D. degrees from Stanford University in 1999,2000, and 2003, respectively, and was an assistant profes-sor 2003–2008 and associate professor 2008–2011 atPrinceton University. His research on networking hasreceived a few awards, including the 2012 IEEE KiyoTomiyasu Award, a 2008 U.S. Presidential Early CareerAward for Scientists and Engineers, several young investi-gator awards, and several paper awards. His inventionsresulted in seven U.S. patents and a few technology trans-fers to commercial adoption, and he received a TechnologyReview TR35 Award in 2007 and founded the PrincetonEDGE Laboratory in 2009. He serves as an IEEE Communi-cations Society Distinguished Lecturer in 2012–2013, andwrote an undergraduate textbook, Networked Life: 20Questions and Answers, for a new undergraduate course,Networks: Friends, Money, and Bytes.

The growing fear of

a data tsunami, par-

ticularly in wireless

networks, has led

ISPs to use pricing as

a congestion man-

agement tool. How-

ever, understanding

the potential benefits

and risks associated

with different pricing

schemes is crucial.

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next PageIEEE

Communications qqM

Mq

qM

MqM

Qmags®THE WORLD’S NEWSSTAND

___

__________________________________________

___________________________________

__________________________________