`Guest Tek v Nomadix, IPR2019-00211
`
`
`
`like are punishable by fine or imprisonment or both under Section 100] of Title 18 of the
`United States Code.
`
`Signed on this 8th day of November 2019, in Houston, Texas.
`
`Edward w. Knightly
`
`
`
`ATTACHMENT 1
`
`
`
`INDEX: A Platform for Determining how People Value the Quality of their Internet Access
`
`Bjorn Rupp* Richard Edell* Harish Chandt Pravin Varaiya”
`*Department of Electrical Engineering & Computer Sciences t Department of Economics
`University of California at Berkeley
`
`Abstract
`
`The continuing exponential growth of the Internet and the emer-
`gence of new time-critical applications have led to the integration
`of a large number of different services on the Internet. In the
`process, the question of how to efficiently allocate bandwidth as
`a scarce resource has become a crucial issue for the continued
`proliferation of these new services. Future growth depends on the
`division of services into quality-differentiated market segments
`and the pricing structure of each segment. Successful growth re-
`quires service providers to o€fer combinations of quality and price
`that match user need. But to do this providers must understand
`the structure of user demand. Such understanding is lacking at
`present.
`This paper describes a platform designed to obtain a basic un-
`derstanding of how individuals value Internet usage when offered
`different Quality of Service choices. The Internet Demand Ex-
`periment (INDEX) has two inain objectives: (a) Measurement of
`user demand for Internet access as a function of Quality of Service
`(QoS), pricing structure, and application; and (b) Demonstration
`of an end-to-end system that provides access to a diverse group of
`users at attractive price-quality combinations. The data being col-
`lected is expected to reveal the correlation between user application
`and service demand, how demand varies with user experience, and
`up to what extent users form discrete market segments. This paper
`gives an overview of both the technology employed at INDEX and
`the goals of the experimental1 design.
`
`1 Motivation
`In recent years, the Internet has undergone a dramatic transforma-
`tion from a computer network dominated by traditional, mostly
`text-based applications and a comparatively small, coherent user
`community to a universal platform for ever more users and ser-
`vices. This was not without its consequences. While traditional ap-
`plications like electronic mail or file transfers can react in an elastic
`fashion to deviations in available bandwidth, new time-critical ap-
`plications like Internet telephony and video conferencing cannot,
`thereby causing their employment to be severely limited as soon as
`network congestion leads to high packet delays and packet drops.
`With the explosion of demand for Internet services, higher speed
`access, and new applications, this situation continues to worsen. A
`single “best effort” service quality seems to become increasingly
`inappropriate for a network serving a wide variety of users and
`applications. Currently, useirs who occasionally need high band-
`width are either forced to lease over-provisioned dedicated lines,
`
`risk the vagaries of the performance of “best effort”-quality shared
`resources, or forego the desired application altogether. When
`demand for Internet access varies among the population (as indi-
`cated by population-projectable data as in [CommerceNetNelsen
`19971 ), quality differentiation, along with proper economic in-
`centives, can increase the overall value of the network by making
`available resources when needed for high value applications. The
`division of services into quality-differentiated market segments
`and the design of appropriate pricing structures for each segment
`is crucial for further proliferation of Internet services. Successful
`growth requires service providers to offer combinations of quality
`and price that match user need. But to do this providers must
`understand the structure of user demand. While there have been
`many pricing proposals in recent literature (for a short overview of
`different approaches, see [Shenker et al. 1996]), such understand-
`ing of user demand is lacking at present.
`INDEX - the Internet Demand Experiment - is a real-world
`market trial seeking to provide this information and measure how
`individuals value Internet usage when they are offered different
`Quality of Service choices. INDEX has two main objectives:
`(a) Measurement of user demand for Internet access as a function
`of quality of service (QoS), pricing structure, and application; and
`(b) Demonstration of an end-to-end system that provides access to
`a diverse group of users at attractive price-quality combinations.
`The experiment will provide Internet access over ISDN lines to a
`group of about 150 users from the Berkeley campus community
`for a two-year period. Users select network services from a menu
`of QoS-price offerings and pay for their usage. It is important
`to stress that while the subjects’ basic Internet access (in partic-
`ular, the ISDN line and access equipment) is greatly subsidized,
`each choice on these QoS menus has a real economic cost which
`the subjects pay out of their own pockets. This is necessary in
`order to achieve incentive compatibility, i.e. given the incentive
`schedule as represented by their active menu, users pick the option
`that corresponds to their true valuation of the networkresources in
`question. The menu changes in certain intervals in order to mea-
`sure demand for a wide range of combinations of QoS, price and
`user characteristics. The data being collected is expected to reveal
`the correlation between user application and service demand, how
`demand varies with user experience, and up to what extent users
`form discrete market segments. The data will also allow to test
`hypotheses about the structure of the market for variable-quality
`ATM services. In addition, the experiment demonstrates a single
`system that offers variable service quality-price combinations that
`meet the needs of a diverse user population, an automated billing
`system that also gives the user control over service selection, and
`
`0-7803-4482-0/98/$10.00 0 1998 IEEE
`
`85
`
`
`
`a remotely operated network monitoring and management system.
`This paper gives an overview of the INDEX Project’s scope and
`describes both the technology employed and the goals, timing and
`structural details of the experimental design.
`
`2 Experimental Setup
`
`2.1 INDEX Access Network Provision
`
`The INDEX access network provides IP service over dedicated,
`128kbps ISDN lines in order to establish a predictable and stable
`QoS between the subjects’ homes and the INDEX Project Net-
`work Operations Center’. For this purpose, INDEX loans a pre-
`configured Cisco 762 ISDN router to each subject participating in
`the experiment and installs an ISDN phone line at their home. The
`128kbps basic rate interface lines coming from the subjects’ homes
`are then multiplexed over ISDN primary rate lines at the Pacific
`Bell central office before they reach the INDEX Project Network
`Operations Center. In contrast to common industry practice, the
`overall available bandwidth is not reduced in the multiplexing
`process and the whole network is heavily overprovisioned to make
`sure that none of the subjects experience deteriorations of their
`selected quality level due to potential bottlenecks at the INDEX
`access network.
`
`..............................
`e
`
`DN BRI
`
`I
`
`I
`
`/ -
`
`2.2 User Interaction, Accounting and Billing
`
`INDEX uses a locally developed system for user interaction and
`metering individual subject usage. The user interacts with this sys-
`tem by means of the “Control Center”, a Java application running
`on the user’s computer. For the subjects, this is the central appli-
`cation enabling them to select different Qualities of Service and
`control their usage of network resources. Apart from functions for
`login and authentication, it consists of a small window informing
`the user about the current experiment, the price schedule currently
`in effect and the actual choices. The subjects can choose a service
`quality by the click of a button and change their Quality of Service
`even during the active session. The Control Center also provides
`usage feedback by displaying a summary of charges for either the
`current session, the current day or the current month.
`
`i
`
`e
`
`I Cisco7200
`
`Figure 1: INDEX Network - Transport Layer
`
`At the INDEX NOC, all connections are through either a Cisco
`7507 or 7513 Internet router. Theserouters distribute all user traffic
`over a set of Billing Gateways specifically designed to meter usage
`and selectively adjust the service quality of individual connections.
`The user may select a service quality from the currently active
`menu of choices at any time. Connections are aggregated by
`user so that the quality for this bundle can then be controlled
`accordingly. All outbound packets are forwarded to a Cisco 7200
`router that is directly connected to the UC Berkeley lOOMbps
`FDDI backbone.
`
`IIt should be noted that although the current experimental setup is oriented
`towards providing service over ISDN lines, the INDEX network architecture is
`flexible enough to allow us to expand the experiment to demonstrate ADSL or
`CATV access using cable modems at a later stage.
`
`Figure 2: INDEX User Interface (“Control Center”)
`
`The Control Center application communicates user choices and
`selected quality levels as control data going through a Billing Gate-
`way to a “supervisor” process. This supervisor process then orders
`the Billing Gateway to treat this user’s connections according to
`the selected quality level. The Billing Gateway in turn meters the
`traffic and reports back to the supervisor process,
`User traffic is monitored and recorded at a fairly detailed level
`for both billing purposes and subsequent offline analysis. The
`database contains records for each TCP connection. Apart from an
`anonymized user ID, time stamp, selected QoS/price information
`and a variety of TCP control data types, they include information
`about connection length, the amount of inbound and outbound
`traffic for the connection, source and destination IP addresses,
`port numbers, and other data describing the type of user activity.
`It is important to collect data at this level of detail in order to not
`only record at what time users change their QoS choices, but also
`to infer what parameters influence these decisions and what the
`reasons for these changes are. Such detailed records are able to
`reveal, for instance, what applications are running at the time of
`a QoS change and what types of hosts and network services are
`involved.
`
`86
`
`
`
`2.3 Network QoS Emulation
`After a subject has chosen a, desired quality level, the QoS must
`be adjusted (i.e. degraded) accordingly. As the current Internet
`infrastructure does not permit controlling QoS, the INDEX Billing
`Gateways do not only control and measure network usage, but
`they are also capable of selectively degrading the performance
`of certain TCP connections (e.g., all connections on behalf of
`a given subject). User quality choices map to entry points of
`an internal “emulated network” composed of different elements
`including leaky bucket, random router and packet delay or packet
`drop, This behavior can be altered quickly in response to subject
`choices or experimentally controlled random processes.
`While it would be desirable to give users complete control over
`end-to-end performance of their connections, this is unfortunately
`impossible while the single-quality “best effort” paradigm still
`prevails. As a consequence, the quality offered by INDEX cannot
`exceed the baseline undegraded QoS. However, much of our sub-
`ject pool is accustomed to a congested, 14.4kbps modem pool for
`accessing the campus network. In addition, many of the services
`that our subjects seek to access are in fact campus services for
`which there are little other sources of degradation. Therefore, we
`believe that INDEX is adequately capable of controlling the QoS
`delivered to the subjects.
`
`3 Experimental Design
`3.1 Purpose and Objectives
`
`INDEX seeks to answer the question of how people value the
`quality of their Internet access. There are three basic aspects to
`this question. The first and most fundamental is what dimensions
`of QoS truly matter for overall end user’s perception of service
`quality. While there are indeed many dimensions of QoS that can
`or could be varied on the supply side, little is known about along
`which dimensions users feel performance to be most significantly
`affected by these parameters and up to what extent they are will-
`ing to sacrifice certain service characteristics in order to improve
`others. Although these internal valuations are very likely to be
`heavily dependent on the type of application the user is running,
`empirical data capable of accurately verifying and quantifying
`these hypotheses is still missing at present. One important goal of
`the experimental design therefore is to identify the key parameters
`for user’s perceptions of Q0S and to quantify the correlation be-
`tween application type and service demand. This information will
`substantially aid network service providers in future decisions on
`which aspects of QoS to optimize and when and where to expand
`the network in order to provide service options that satisfy users’
`needs.
`The second task is to measure the economic value which indi-
`vidual users place on different resource levels for each of the QoS
`dimensions identiJied. The results from these investigations should
`support decisions on what kind of pricing structure to choose and
`what economic incentives the price structure should provide. For
`example, congestion-based prices require users and network ad-
`ministrators to know the value lost due to congestion. The degree
`to which congestion needs to be discouraged or the markup charged
`
`to priority service depends on the degradation of user value due
`to user contention. If the perceived network degradation from
`congestion is substantial, then congestion-related pricing (such as
`time-of-day, traffic-based, or priority-based charges) can rational-
`ize the allocation of network resources and increase the value of
`the network.
`Apart from gaining a better understanding of how to quan-
`tify valuation of different dimensions of QoS, survey data (La.
`[CommerceNetMielsen 1997, Kehoe/Pitkow/Morton 19971 ) also
`suggests that demand for Internet access varies among the popu-
`lation. The data also indicates that there is a correlation between
`user experience and intensity of networkusage. If users are indeed
`considerably heterogeneous in their consumption of network re-
`sources, information about the exact nature of the price elasticity
`of demand will help to differentiate types and levels of service
`through pricing. Therefore, INDEX also includes experiments
`involving nonlinear tariffs to determine up to what extent users
`form discrete market segments. Such multi-part tariffs involve
`price discrimination in the sense that different bundles of homo-
`geneous output are sold at different prices. If substantial variation
`exists among users, and if users are sufficiently sophisticated in
`their decisions, then self-selecting tariffs differentiated on the ba-
`sis of quantity and quality may segment the market on the basis of
`willingness to pay.
`
`3.2 Subject Population
`
`INDEX will recruit about 150 subjects affiliated with the Univer-
`sity of California at Berkeley (students, faculty, staff). Participa-
`tion in the Project is highly attractive for this group of experimental
`subjects because the campus modem pools are highly congested
`while there exists at the same time a lack of other options offering
`service similar to INDEX. The availability of this subject pool of-
`fers several advantages: Firstly, it provides the required geograph-
`ically concentrated pool of diverse users for such an experiment.
`It also allows for a better QoS control than otherwise possible,
`ensures continued participation and reduces costs of setting up the
`experiment.
`The recruiting process involves several steps: The project ad-
`vertises by electronic means such as posting to newsgroups as
`well as traditional means of Berkeley newspaper advertisements
`and articles. The advertisements direct interested people to the
`INDEX WWW server to learn more about the project. Potential
`subjects who wish to participate are required to complete an on-
`line screening survey. The INDEX screening survey collects basic
`contact information, residence location and nature of university
`affiliation to verify eligibility and aid in ISDN service planning.
`Prospective subjects must reasonably expect that their affiliation
`with UC Berkeley will continue for at least two years. We also
`plan to facilitate participation from individuals affiliated with UC
`San Francisco in the near future. After prospective subjects have
`completed the screening survey, they are invited to complete an
`extensive demographic survey. At this stage, taking the survey
`is optional, however completing it is required for all participating
`subjects before they take part in the experiment. After evaluating
`the screening survey results, we invite selected persons to par-
`ticipate. If they agree, they receive their access equipment and
`
`87
`
`
`
`can begin using their INDEX-provided Internet access after the
`ISDN line has been installed at their home and they have signed
`an informed consent form.
`Subjects are recruited with the goal of obtaining a suitable
`variation in, e.g., field of study, expected computer usage, travel
`distance to campus, and demographic characteristics. Neverthe-
`less, it is evident that the sample, like any other, remains biased. In
`order to overcome this sampling bias and be able to extrapolate the
`results, significant sample-specific characteristics influencing the
`results will be identified and analyzed to determine the exact struc-
`ture of the INDEX Project’s demographic base. Before receiving
`INDEX provided Internet access, each subject has to complete
`a detailed demographic survey. Many of its questions are taken
`from a representative, population-projectable study conducted by
`Nielsen Media Research [CommerceNetMielsen 19971 in early
`1997. Apart from general demographical data (i.e. income, age,
`gender, household characteristics etc.), the INDEX Demographic
`Survey asks subjects about their recent Internet usage, computer
`sophistication and related data. By comparing the responses to the
`INDEX Demographic Survey with the Nielsen data, it is possible
`to extrapolate findings from the INDEX Project to the general U.S.
`population2.
`
`3.3 Revealed Preference Experiments
`
`To achieve the objectives outlined above and investigate the effects
`of alternative price and QoS combinations, a series of experiments
`will be conducted. These experiments infer preferences from data
`based on actual choices for which the subjects have to make eco-
`nomic decisions immediately affecting them - therefore providing
`reliable incentives to accurately represent their preferences. The
`experiments are preceded by a free trial period of several weeks.
`This serves as a “control” for later INDEX experiments and al-
`lows the subjects to gain familiarity with their new Internet access,
`the billing interface and experimental procedures. After the trial
`period is over, users begin paying for their usage. Depending on
`the experimental setting, the fee structure will change either every
`Monday on a weekly basis or daily at 4 a.m. Pacific Standard Time.
`Higher qualities of service will incur a higher fee, but the level and
`ratios depend on the specific sub-experiment and will also change
`over the course of each experiment. Whereas nonexperimental
`studies are forced to rely on cross-sectional variation in prices and
`demand to infer the price elasticity, varying the prices during the
`individual sub-experiments allows for measuring the demand re-
`sponse for each participant. Fairly detailed data will be collected
`on the characteristics of the subjects’ usage. As described in sec-
`tion 2.2, each time the subject uses the ISDN connection, data will
`be collected on the time and length of the session, the speed of the
`connection, the price in effect, and the amount of data transferred
`and applications used during the session.
`The experiments can be partitioned into two groups. The first
`group, representing the first four experiments, examines how the
`zWe also plan to expand the experiment to conduct general population exper-
`iments and test CATV and ADSL access in the future. This is dependent upon
`cooperative arrangements with CATV and ADSL service providers which have not
`yet been negotiated. These future experiments will be based on a different ISP
`Resource Model and are not describedin this paper.
`
`alii >Gp-----FkQ
`
`:I
`Subscriber
`LOOPS (N)
`
`I-
`
`Access
`Ports (M)
`
`ISP Link
`
`Packet Switched
`Circuit Switched
`(B bps total)
`(A bps per circuit)
`Figure 3: Traditional ISP Resource Model
`
`dimension of resources within the traditional ISP resource model
`(as depicted in Figure 3) affect end user valuation. The second set
`of experiments, experiments five through nine, examine users’ re-
`sponse to alternative price structures. In the majority of these later
`experiments, users will compare alternative pricing schemes with
`a flat rate scheme. The following sections describe the individual
`sub-experiments in detail.
`
`3.3.1 Network Resource Valuation Experiments
`(1) -Variable Bandwidth. Do individuals value connection speed
`sufficiently to pay higher prices for high speed connections? How
`does the elasticity of demand depend on application and demo-
`graphics? Does demand exhibit habit formation? This experiment
`will address these questions and examine how users value the
`speed of their connection to the Internet cloud. It isolates the last
`link in Figure 3 and permits the accurate measurement of the price
`elasticity of demand for connection speed to the Internet cloud.
`Design Details. QoS dimension controlled: Bandwidth (six connec-
`tionspeedsA = {8,16,32,64,96,128kbps}). Duration: 6weeks. Price
`structure: Five weeks with weekly price changes, one week with daily
`price changes - strictly increasing prices within certain limits.
`(2) - Variable Asymmetric Bandwidth. Since the early days
`of the 1200/75 bps modems, the question of whether individual end
`users value bandwidth for incoming traffic more than for outgoing
`traffic has been discussed intensely. With the advent of ADSL
`and CATV proposals that feature different fixed data rates for
`incoming and outgoing traffic, this issue warrants further research.
`Do individuals value the speed of their connection from the Internet
`cloud differently than the connection speed to the Internet cloud?
`If yes, what ratios of incoming vs. outgoing bandwidth are deemed
`appropriate? Up to what extent do these ratios depend on the type
`of application run by the user? This experiment seeks to answer
`these questions.
`Design Details. QoS dimension controlled: Bandwidth (six connec-
`tion speeds for both incoming and outgoing traffic). Duration: 6 weeks.
`Price structure: Five weeks with weekly price changes, one week with
`daily price changes - prices from the first experiment will be cut in half
`(3) - Access Reliability. Congestion in dialup access to the
`and applied separately towards each direction of all data flows.
`network is a significant problem in many networks. The expected
`waiting time for a free line can be significant. As a result, once
`
`88
`
`
`
`connected, users are typically reluctant to relinquish their connec-
`tion even during idle periods for fear of inability to reconnect in
`a prompt manner. By presenting subjects with a choice of three
`different simulated modem pools, this experiment will examine
`the value which users attach to network access and the value of
`waiting time. Each of these simulated modem pools is associated
`with a different, predetermined level of congestion and average
`waiting time. The experiment will thus provide data on whether
`individuals value network access sufficiently to pay higher prices
`for less contention for free circuits and whether they relinquish
`lines more often when they can reconnect.
`Design Details. QoS dimensions controlled: Expected waiting time
`for a connection, bandwidth. Duration: 4 weeks. Price structure: Fixed
`fee based on choice of average modem pool congestion changes daily,
`per-minute charge based on connection speed will not change during the
`duration of the experiment.
`
`(4) - Priority Service. Another potential source of contention
`from other users is for shared bandwidth (labeled B in Figure 3)
`such as in trunk lines or Internet cable access. While the last ex-
`periment addressed the issue of access reliability, this experiment
`will measure the value which users place on service reliability and
`effective bandwidth by measuring their willingness to pay for dif-
`ferent service priorities. The level of effective bandwidth which
`subjects obtain will be a function of the chosen priority class and
`the level of congestion in the network. The interfering traffic will
`be simulated. The findings from this experiment are expected
`to yield insights into the possible future role for multiple service
`classes that take into account the load status of the network nodes.
`Design Details. QoS dimensions controlled: Effective bandwidth,
`dependent on varying amounts of simulated interfering traffic. Duration:
`6 weeks. Price structure: Different prices for different priority classes.
`
`3.3.2 Alternative Pricing Structures Experiments
`(5) - Demand under TraMis-Based Charges. To the extent that
`service degradation occurs due to network traffic, prices should
`reflect the level of traffic which the user generates. However,
`charges based solely on connection time do not provide the proper
`economic incentives and result in a negative externality and in-
`creased congestion. While many researchers have suggested that
`users should be charged on the basis of their traffic, little is known
`about whether users will adequately understand the basis for such
`prices, whether usage will be sensitive to such prices, and whether
`users could benefit from usage based prices due to a reduction in
`network congestion: How price sensitive is the amount of user
`generated traffic? What are the optimal usage charges for band-
`width on demand? How much learning about traffic generation
`occurs? Does the price elasticity change over time?
`Design Details. Segmentation Approach: Volume. Duration: 8 weeks.
`Price structure: Per-byte charges which change once per week. Prices
`will be calibrated so that total charges would be approximately the same as
`under per minute charges in the absenceof behavioral effects. Of primary
`interest is whether usage decreases in comparison to the first experiment,
`(6) - Tariff Self-Selection.
`and whether usage decreases more when per byte charges are higher.
`Would individuals voluntarily
`choose tariffs based on bytes rather than minutes? Are these
`choices of self-selecting tariffs rational? Is usage sufficiently pre-
`
`dictable? In this experiment, subjects will be presented with a
`choice of six different connection speeds and three different tariff
`structures. The tariff structures will be different combinations of
`per minute charges and per byte charges. Users will be offered the
`choice of being charged exclusively on the basis of either minutes
`or bytes, as well as the choice of having half of the charges being
`determined by each.
`Design Details. Segmentation Approach: Self-selection. Duration:
`10 weeks. Price structure: Users will be asked to select which tariff
`structure they wish to be billed under for the course of the week. The
`prices contained in the tariff structure will remain in effect for two weeks
`at a time.
`(7) - Time-of-Use Charges and Peak Shifting. Network use
`displays regular temporal patterns. In order to avoid building extra
`capacity to meet peak usage, price incentives may be provided in an
`effort to shift some demand from peak to non-peak periods. Since
`there is likely to be a significant amount of variation in tastes
`and time constraints, significant welfare gains might be possible.
`For example, if a sizable segment of the user population could be
`induced to shift usage to nonpeak hours, the value of the resources
`to those unable to shift their usage may increase substantially.
`Design Details. Segmentation Approach: Time-of-day. Duration: 8
`weeks. Price structure: Price incentives will be provided to shift some
`demand from peak to non-peak periods. Suitable peak periods will be
`identified from data collected during the earlier experiments.
`(8) - Demand under Flat Rate Pricing. This experiment
`will vary the level of the flat fee charged to users for unlimited
`usage in order to examine the total value which users place on
`access at different connection speeds. The relationship of this
`value to various observable characteristics will also be examined.
`By obtaining this measure of user heterogeneity in the total value
`of usage, the design of optimal tariffs will be helped greatly.
`Design Details. Segmentation Approach: Total value of access at
`defined connection speeds. Duration: 12 weeks. Price structure: Flat
`price will differ by connection speed and change on a weekly basis. Unlike
`the first experiment where each connection speed could be reselected as
`desired, only the speed@) for which the flat fee has been paid will be
`allowed. Thus, if an individual discovers in midweek that a higher speed
`is needed, the individual would have to pay the new higher fixed fee.
`
`(9) - -0-Part Tariff Design. This experiment will com-
`bine some of the different features of the previous experiments.
`Whereas experiment 6 examined the selection of tariffs which
`varied according to the basis for usage sensitive fees (per-minute
`charges versus per-byte charges), this experiment will vary only
`the relative contributions of the fixed and variable charges. This
`will allow examination of how users respond to declining block
`tariffs and also provide some insight into whether individuals have
`preferences over tariff structures themselves due to such effects
`as risk aversion or mental accounting costs. Combined with data
`from the other experiments, this will allow for the design of price
`structures which track consumer demand curves more closely than
`do either fixed fees or uniform prices.
`Design Details. Segmentation Approach: Self-Selection. Duration:
`10 weeks. Price structure: Individuals will face a menu of tariff choices
`which will vary according to the level of the fixed weekly fee and the
`usage sensitive fee and change each week.
`
`
`
`users. INDEX will provide much needed real-world data on an
`increasingly important issue capable of transforming the Internet
`in a universal platform for ever more users and services.
`Another issue of interest in regard to the introduction of quality-
`differentiated service is that decisions to be made by the user will
`become increasingly complex. Although some existing QoS archi- ,
`tectures shield applications from the complexity of the underlying
`QoS organization, the problem of how user preferences regarding
`prices for certain service qualities can be incorporated in these
`mechanisms has not yet been adequately addressed. User sup-
`port for these tasks might come from a program running on the
`user’s computer that analyzes user traffic in order to get informa-
`tion about the network performance and the kinds of application
`currently running. Such a program could then make appropriate
`purchasing recommendations or even conduct these transactions
`on its own once the user has specified his price and quality pref-
`erences. INDEX will also examine these issues, and work is
`currently underway to model them and develop a software proto-
`type.
`References
`
`[ 11 CommerceNetNielsen Media Research (1997): The Com-
`merceNeiYNiefsen Media Research Spring 1997 Internet De-
`mographic Survey. Full Report. New York: Nielsen Media
`Research, 1997
`[2] Edell, Richard; McKeown, Nick; Varaiya, Pravin (1995):
`Billing Users and Pricing for TCP. In: IEEE Journal on
`Selected Areas in Communications, vol. 13 (7), pp. 1162-75
`[3] Kehoe, Colleen