Dynamic Pricing for 3G Networks Using Admission
Control and Traffic Differentiation
Vitalis G. Ozianyi and Neco Ventura
Department of Electrical Engineering
University of Cape Town
Rondebosch, WC 7700 South Africa
Abstract-In the pricing of network resources, network op-
erators and service providers aim at facilitating the use of the
limited network resources in a manner that would encourage
responsibility among the end-users and lead to the maximisation
of profits. The optimum tariff rates used for charging the mobile
services are affected by factors like the market forces affecting
the industry. However, the tariff rates generally increase with the
achieved QoS level. Next generation networks will offer higher
QoS, hence users need incentives to utilise the enhanced capacity.
In this paper, we propose a pricing approach that introduces
service profiles into a DiffServ-enabled network, whose prices and
QoS levels depend on the degree of congestion in the network.
The use of the UMTS connection admission control to support
the proposed pricing scheme is explored. An emulation testbed is
used to evaluate the scheme.
Indecx Terms-Pricing, Quality of Service (QoS), Class of
Service (CoS), Differentiated Services (DiffServ), PDP context, 3rd
generation (3G).
I. INTRODUCTION
T he provision of mobile communication services requires
installation of expensive network equipment that need
constant maintenance to guarantee the availability of services to
end-users at all times. The demand for network services has in-
creased rapidly, thus leading to the need for proper management
of the limited resources. Pricing is used for controlling the use
of network resources, and it facilitates charging that enables
the network operators/service providers to generate revenue
for offsetting their capital and operating expenses (CAPEX
and OPEX). The prices for the services are reflected in the
applicable tariff rates, and they determine the market advantage
gained over other operators in the region. This calls for market
surveys [2], [3] that are targeted at the interests of the end-user.
3G networks offer new services (e.g., real-time multimedia)
and they are designed to offer guaranteed (QoS) together
with increased levels of resource availability [4]. Multimedia
services require guaranteed network conditions e.g., low delay
and jitter [5]. Some Internet services (e.g., e-mail access,
web browsing etc.,) can tolerate unstable network performance
conditions. In the General Packet Radio Service (GPRS) and
the Universal Mobile Telecommunication System (UMTS),
The authors would like to thank Telkom SA, Siemens, the National Research
Foundation (NRF) and the Department of Trade and Industry (DT[) for
supporting this research project
IRefer to [Il for the definition of terms related to QoS
network resources are defined using QoS parameters, whose
values are specified during the initialisation or modification of
a packet data protocol (PDP) context [6]. Although the tariff
rate used for charging a given PDP session depends on factors
that are service specific, the applicable QoS parameter values
would have a substantial influence [71.
The individual services offered on mobile networks differ
in the amount of resources that are required for successful
delivery. Congestion is always directly proportional to' the
demand of network resources at a given time, hence when
the number of PDP sessions that are being served increases,
congestion is bound to occur. Considering the amount of
bandwidth and other resources that are specified in the design of
3G and future generation networks, a PDP session can request
and be allocated QoS resources that could be a multiple number
of times greater than the levels achievable on 2G networks.
End-users will not get incentives to use some mobile services
(e.g., multimedia applications) if the current tariff structures
are used, due the possibility of incurring high charges. There
are periods when 3G networks are bound to have a surplus
resources. This paper proposes a criterion that could be used to
authorise the extra resources, at a discounted tariff rate, to user
applications that request them. Since this will lead to dynamic
prices and QoS levels, users need to indicate, in real-time, their
willingness to continue using the services at new tariff rates and
QoS levels.
The introduction of service profiles will allow users to
specify, prior to using the network, their service usage profile
and in tum influence the QoS vs. cost relationship of the
services delivered to them. In this paper, a platinum profile that
guarantees high QoS, a constant-tariff rate gold profile, and a
flat-rate priced silver profile that offers best effort service are
explored. This architecture is designed for IP services in 3G
networks, in which the DifServ- architecture [8] is used for
QoS provisioning.
The rest of this paper is organised as follows; Section II
gives a review of pricing and QoS in mobile communication
networks, Section III presents the architectural design of our
proposal, the design and implementation of the testbed are
given is section IV and V respectively, section VI presents the
evaluation results, and section VII concludes the paper.
1-4244-0000-7/05/$20.00 C2005 IEEE. 838
II. PRICING AND QoS IN MOBILE NETWORKS
The evolution of pricing and charging schemes for telecom-
munication and Intemet services is a continuous process.
Many pricing proposals have been made, yet a few have
been implemented in commercial systems. In the selection
of pricing schemes, network operators consider factors like
simplicity, scalability, the signalling and processing overhead,
and the feasibility of implementation [9]. On the other hand,
the users' requirements for a pricing/charging2 scheme include:
predictability of charges, accuracy and convenience [111. In
terms of simplicity, the flat-rate pricing [121 scheme is at
the top of the list, and it is popularly associated with the
Internet. In flat-rate pricing, the charges do not reflect the
level of network resource usage, which leads to unfairness
between heavy and light utility users and also loss of rev-
enue for operators when service usage is above the estimated
average levels. The usage-based and special pricing schemes
that have been proposed include, Paris Metro Pricing (PMP)
[13], which creates differentially priced channels with better
service expected in the higher priced channels due to less
usage, and the responsive pricing, which exploits the adaptive
nature of users to improve economic and network efficiency
by increasing network prices during periods of congestion [3].
Other interesting pricing proposals are priority pricing, edge
pricing, and effective bandwidth pricing.
Charging enables network operators to generate revenue from
the services offered. The charging approach used is operator
specific; however, it depends on the pricing scheme. A common
trend in telecommunication charging relies on the time-of-day
(ToD) pnrcing [14], which is a version of the responsive pricing
scheme. In ToD, network congestion is assumed to be higher
during daytime than nighttime, hence the former is priced
higher than the latter. It is clear that pricing and congestion con-
trol are closely linked. All applications have minimum network
performance requirements, hence most networks are designed
to meet these requirements. QoS provisioning in IP networks
can be achieved using two architectures; the integrated services
(IntServ) [15] and the differentiated services (DiffServ) [8].
The DiffServ architecture creates classes of service (CoS), to
which different applications are grouped. Packets belonging to
different CoS would receive a characteristic network perfor-
mance treatment called per-hop-behaviour (PHB). The 3GPP
has proposed standards for an all-IP mobile network. This
makes it possible to use DiffServ for QoS provisioning in
mobile networks [16).
Since DiffServ does not guarantee QoS for individual flows
in a CoS, the use of connection admission control facilitates the
reservation of QoS resources the active flows. In the UMTS,
the amount of resources needed to support a PDP context
are specified in terms of QoS parameters. For packet data
communication, control of the use of network resources, in
the core network (CN) and the radio access network (RAN), is
done by the gateway GPRS support node (GGSN), the policy
2Refer to [10] for a definition of terms related to billing
Fig. 1. UMTS QoS signalling flows.
decision function (PDF) and the application function (AF) [6].
A PDP session is identified by the combination of the source
and destination IP addresses and port numbers. Within the
UMTS CN, the MSISDN (Mobile Station ISDN) and the IMSI
(Intemational Mobile Subscriber Identity) are also used for
session identification [4].
When the user equipment (UE) activates a PDP context,
authorisation of UMTS resources for the up-link and down-
link directions would be done if the network has the resources
needed to support the request. Downgrading of the requested
UMTS QoS parameters may be done when the available
network resources are less than the required amount. Fig.1
illustrates the flow of information during the PDP context
activation3. The information needed for charging the session,
e.g. the applicable tariff rate is exchanged during this procedure.
Individual operators decide on the exact monetary values to use
for the tariffs. However they rely on factors like market forces
affecting the mobile industry, and their aim is to offset the
CAPEX and OPEX and to generate profits. From a technical
view point, the QoS authorised for the bearer affects the
applicable tariff rate [18].
Accounting of network resource usage is required in usage-
based charging. Charging detail records (CDR) are formated
files that contain information on the service and the user to
whom it was delivered. They are generated by the SGSN
and the GGSN [181, and they are used in billing for the
use of the network. Whereas flat-rate charging only attracts a
periodical subscription charge, usage-based charging comprises
a subscription charge and a per-connection charge. The per-
connection (access) charge can be expressed in an abc format;
aV + bT + c (1)
where T and V are the measured duration and volume of
the connection respectively, and a,b,c are tariff parameters
applying to the connection.
The data rates available in mobile networks increase with
each generation. 2.5G networks offered up to 144kbps, while
3G networks are designed to provide up to 2Mbps. Future
generation networks will have higher data rates, since inter-
working between 3G and wireless local area networks (WLAN)
creates access networks with more than 10Mbps [19]. The
enhanced data rates will facilitate the authorisation of high QoS
3Full details for information flow in Fig. I are available in [171
839
UE
r7.--
.--4---
- M..-
t
t
VOATS 86 &A-g-
AP SW-wg
---
1 6..P%..*- F.."-)I
GGSN
Gq
C-
(-.-.",--)
-=-+--- -,
. ho
-
I
I
- t ---
--- #4 ..0-
ZO"r-g P."
for network services and applications. Using current mobile
tariff structures, authorisation of higher QoS would lead to
incurring charges that could be hundreds of times the values
charged currently.
It is worth noting that very high charges will fail to provide
incentives for the utilisation of the full capacity of the network,
which is necessary for improving the efficiency of the network
and user satisfaction. These are the key features that would
enable network operators to generate more revenue. Section III
presents an architecture for pricing network resources during
periods when congestion is high and when the network has an
abundance of resources. The use of service profiles that would
enable users to influence the QoS vs. cost relationship of the
services is investigated.
A Network service profiles
Three profiles are used in this architecture; however more
profiles could be used. Rtp represents the amount of re-
sources allocated to a given profile. If the number of active
flows/sessions in a the profile is Ip, the amount of resources
allocated to each flow would be;
Rtp
tp = IP (3)
This fonnula assumes that each flow in the profile is allocated
an equal amount of network resources. When CAC is used to
limit the maximum number of sessions in a profile to Np, each
flow would be guaranteed a minimum amount of resources that
is equivalent to;
III. ARCHITECTURAL DESIGN OF THE PRICING SCHEME
The Dynamic QoS-based Charging Model (DQBCM), as
proposed in this paper, utilises the four UMTS QoS classes that
have been standardised by the 3GPP [5], [20]. Considering the
limited resources in the access networks, we model a network
with a total amount of resources Rt. These resources are shared
by traffic in the different CoS of the system. The distribution
of network resources to the individual CoS would be operator
specific. In this architecture, the conversational and streaming
multimedia CoS are combined to form one multimedia CoS,
hence the system uses three CoS, i.e. multimedia, interactive
and background. Flows in the multimedia CoS require ex-
pedited forwarding and minimum jitter, while data integrity
(i.e assured packet forwarding) is the main requirement of
the interactive and background CoS. Applying the formula in
eq. 2, the multimedia CoS will receive a bigger proportion
of Qt, which corresponds to higher queue priority, while
the interactive and the background CoS will receive a larger
proportion of bt. Here bt represents the buffer capacity at each
node along the end-to-end path. Network bandwidth (Bt) is
required by traffic in each CoS.
n=n
Rt=E(Bt+bt+Qt) (2)
n=1
Flexibility is introduced into the system through network
service profile that enable the users to specify, before hand,
their QoS requirements. With the introduction of admission
control, each CoS can guarantee QoS from a minimum to
a maximum level, depending on the number of active flows.
The introduction of profiles into each CoS partitions the QoS
range into portions that can be used to enable the users to
optimise QoS for specific services. The fragmentation of each
CoS creates profiles that are priced differently, with the profiles
offering higher QoS levels being priced higher as compared to
the others. When users select a given profile, they enter into a
service level agreement (SLA) with the network operator, and at
the same time they give indication to the network management
system on their willingness to pay for services at conditions
specific to the profile.
(4)Np
The value of rip will be equal to rmp when Ip - Np, i.e when
the full capacity of the profile is used.
Platinum profile: This profile offers services at the highest
QoS level in any CoS. When congestion occurs in the network,
the QoS received by flows in this profile is guaranteed. In this
architecture, the number of active sessions is used to quantify
the demand for network resources, which also relates to the
degree of network congestion. The tariff rate for charging users
of this profile is dynamic, and it varies with the number of
active sessions, i.e. t oc f The tariffrate reaches its maximum
when Ip. Np. We dehne a value of Ip, where Ip = S,
below which the tariff rate would remain constant, thus the
incremental tariff rate for the platinum profile is ti = Kp( I s )
where ,p is a tariff constant for the platinum profile. By setiing
a minimum tariff threshold, the operators are guaranteed of
revenue during periods of light network utilisation. Lowering
tariffs encourages users to use to utilise the network during
periods of less congestion, hence the overall network efficiency
is improved. Raising of tariffs during congestion periods helps
in the control of congestion. Since users are not able to predict
tariff changes, this profile targets users who are less sensitive to
network price changes, and whose main aim is to get services
at guaranteed high QoS level.
Gold profile: The gold profile is designed for users who are
very sensitive to changes in network prices, and also want to
receive services with some level of QoS guarantee. The tariff
rates are not affected by the demand of network resources,
hence they remain constant. The cost of transmitting, say a
multimedia message (MMS) of a certain size remains the same
over periods of different congestion levels. There is a trade-
off between predictable tariff rates and the achievable QoS,
i.e. the QoS offered by this profile is affected by the demand
for network resources. The QoS will be at the minimum when
Np= Np, as illustrated by eq. 4. Using the thresholds defined
for the platinum profile, when Ip - S, the QoS received by
each flow in this profile will be at the maximum, and the
decremental change in QoS as the number of active sessions
increase is given by qi = c9 ( sN ). Since the maximum QoS
840
resource value for each flow in the gold profile (RP) is equal to
the resource allocations for each flow in the platinum profile,
the tariff rates used for the gold profile will be equal to the
minimum tariff rates of the platinum profile.
Silver profile: The silver profile suits users whose main
aim is to remain connected to the network for long durations.
This profile is characterised by the lack of QoS guarantee
and flat-rate pricing. Some applications in the interactive and
background CoS, which are adaptive to substantial changes
in QoS, would be appropriate for this profile. The best effort
transport is appropriate for this profile; hence it is considered
unsuitable for the multimedia CoS, since the minimum QoS
requirement would not be met [5]. The CAC function limits the
maximum number of flows that can be admitted to this profile,
thus ensuring that the active sessions do not suffer breakdown
as a result of excess congestion.
B. Bandwidth sharing between profiles
The architectural design of the DQBCM scheme allows
resource sharing between service profiles. Resource sharing is
done when the demand of resources in a given profile is less
than its capacity, and it is achieved by allocating the excess
bandwidth to a profile where the demand for resources is high.
The sharing of resources can be done between profiles of
different CoS. When, for instance, the gold profile gets excess
bandwidth, the QoS received by its flows can be increased,
or an extra session can be admitted, which makes Ip > Np.
This condition allows the operators to maximise revenue by
shifting excess resources to profiles where demand is high.
The platinum profile would benefit from resource sharing,
since the extra sessions will be charged at the peak tariff
rate. The following condition must be met for resources to
be shifted from the gold profile to the platinum profile; the
excess resources in the gold profile should be at least equal to
the amount of resources required to support one session in the
platinum profile. Referring to eq. 3 and 4, eq. 5 illustrates this
requirement;
(ripg - rmpg) * Npg
f is a constant representing the ratio of resource allocations
in the platinum profile to the minimum resource allocations
in the gold profile, i.e. f = ripg and rmpg are the
respective current and minimum resource allocations for the
gold profile, while rmpp represents the resource requirements
for each platinum flow. Npg is the maximum number of
admissible sessions for the gold profile.
C. Admission control strategy
Admission control is done so as to limit the number of
sessions that are served in a given profile. With reference to
the UMTS network, policy-based admission control is done by
the GGSN in conjunction with the PDF and the AF (refer to fig.
1). In the DQBCM architecture, the admission control function
Databae
-____--- _ ag
AdfsUn 4= AM"MWg L En."
F.4. (ACF) Ad.#&W0-~gF.~
Legened.
Data flow
Sitgnalfing
Fig. 2- DQBCM network management architecture
(ACF) keeps track of the number of active sessions (Ip) in each
profile and the resource allocations for each session. In the
UMTS, the total amount of resources allocated in a profile will
be given by the sumn of the values of the authorised PDP context
parameters of all the sessions. The ACF will authorise new
session establishment attempts when I, < Np1 in the profile.
Fig. 2 illustrates the network management architecture of the
DQBCM scheme.
IV. DESIGN OF THE EMULATION TESTBED
Various functions of the DQBCM architecture can be evalu-
ated on a testbed. This section presents the design of the testbed
used in. performing the evaluations. Classes of service (CoS)
are defined based on the network performnance requirements
of the traffic
-they would handle. Packets belonging to a given
CoS can be' identified using the type of service (ToS) byte
in each packet. Peer-to-server (p2s) and server-to-server (s2s)
communications generally use standard port numbers that are
issued by the Internet Assigned Numbers Authority (LANA),
hence classifying their traffic is simpler, and can be done
in a static fashion; however, peer-to-peer (p2p) sessions -use
ports that are randomly assigned by the end nodes. For this
reason, the classification of traffic from p2p applications must
be done dynamically. The user profiles are stored in a database
on the network management system (refer to fig.2), and will
be matched against the user's static network ID. The network
access identifier (NAI) for each user equipment (UE) is dynam-
ically assigned. Mapping of the NAI to the network ID makes
it possible to identify the service profile that is applicable to
the user.
Tphe UEs will be mobile and their access to the network
can be achieved by configuring a media access gateway (AG).
The AGs would be remotely located, and a network can have
many AGs that are managed centrally by a network access
controlnler (AC). The service profiles selected by the users, plus
other user information should be stored in a database close
to the AC. The AC would handle the connection admission
control functions, traffic management and tariff generation, as
illustrated in fig. 2. The DQBCM system manages the radio
access network resources (e.g., the total bandwidth Be). Bt is a
limited resource that would be allocated to the active flows that
are admitted to each profile. Since the traffic control is partly
based on the DiffServ architecture [81, the assumption made is
841
Access point
Fig. 3. Testbed layout
that traffic is handled according to the appropriate PHBs on the
external networks. The main entities of the network system are
as follows:
User Equipment: The user equipment (UE) enables the
user to connect to the network, and also provides the profile
modification interface. The interface can be menu-driven or
web-based.
Media Access Gateway: The media access gateway (AG)
assigns a NAI to the UE, and enforces the network access
control decisions that are made by the AC.
Network Access Controller: The network access controller
(AC) handles the network management functions, i.e. authen-
tication, authorisation, user profile management etc. These are
either enforced within the AC, as shown in fig. 2, or the AC
authorises another network entity e.g. the AG to perform the
function. The AC performs user authentication and authorisa-
tion (AA) upon receiving a request from the AG to which the
UE is connected. AA is done by querying the user database
for user identities. The AC performs a two-step authorisation
procedure before granting services to the user. In the first
step, the UE gets authorisation to access the network, and
initiates requests for services. The second authorisation step
is performed upon receiving service requests from the user.
This step would only succeed if the user profile in which the
requested service falls has sufficient network resources; this is
done done by the admission control function. Tariff generation
for the platinum profile is done as described in section III-A.
V. IMPLEMENTATION OF THE TESTBED
A testbed for evaluating the DQBCM scheme was setup on
a WLAN. The AG and the AC were implemented on Linux
PCs. Three Classes of service (CoS) were created based on
standard port numbers of network applications (e.g., port 25 for
SMTP and port 80 for HTTP traffic). Details on implementing
DiftServ on Linux may be found in [211, [22]. The profile
selected by each user was identified by mapping the NAI (i.e.
UE's IP address) to the corresponding profile entry that matched
the user's unique network ID (i.e. UE's MAC address) in the
profiles database. The UEs were emulated using laptops with
wireless connection to an access point, as shown in fig 3.
The various entities of the testbed worked as follows: each
UE was issued with a NAI by a dynamic host configuration
protocol (DHCP) process on the AG. The AG transmitted the
NAI and UE's MAC address to the AC for processing. On
successful authentication, the AG was instructed to perform the
first authorisation step. This was done by adding the NAI and
the MAC address to an IP-TABLES firewall list on the AG. The
second authorisation step was evaluated using TCP traffic. The
TCP connection setup process uses SYN packets, which were
only permitted through the AC's traffic control system when
Ip < N2p for the profile affected by the request. Admission
control for TCP traffic was achieved using the connection
tracking module that is available in Linux [221, and which
was used to count the number of active connections (Ip) in
each profile. Connections were uniquely identified using their
source/destination IP address and port combinations.
In conducting a performance evaluation of the QoS mech-
anism, UDP traffic generators [231 were introduced. The AG
was replaced by three traffic generators, which sent traffic to
a host node located on the external networks side (refer to fig.
3).
VI. EVALUATION AND RESULTS
Timing data for the authentication and authorisation (AA) of
the UE were as follows; the average time between detecting
the UE and authorising it to use the network was 0.214s,
which included the average time between initiating an AA
request by the AG and receiving a response from the AC, which
was 0.044s. These delay times are small and would not have
diverse effects on the performance of other network related
procedures e.g. handoffs between networks. The connection
admission control function achieved 100% efficiency under
TCP traffic; however, it could not be tested with UDP traffic.
In mobile networks, admission control algorithms are quite
advanced (e.g., as illustrated in section II). The performance
of the DQBCM scheme when tested with traffic generators is
given by fig.4 and 5. Fig.4 compares an over-provisioned and
a typical network, where the former gives better performance
than the latter. In the second set of graphs in fig.4, the real-
time CoS is allocated higher bandwidth (BW) and a higher
queue priority than the interactive and background CoS, with
the background CoS getting minimum BW and lowest queue
priority. The real-time CoS depicts a better performance index
as congestion sets in. With bandwidth sharing, the performance
index of the interactive and background CoS improves, before
congestion sets in. This is because some of the unused BW
from the real-time CoS is allocated to them hence boosting
their throughput. In fig.5, the platinum, gold and silver profiles
have same queue priority, but the BW allocation is greater for
the platinum CoS and least for the background CoS. The BW
sharing effect also applies to this case.
A. Comparative evaluation of the DQBCM
Using the evaluation criteria given in [3], the DQBCM
scheme compares as follows: it is compliant with IP networks,
hence it targets 3G and next generation networks (NGN)-
Network resources are needed for resource usage accounting in
the gold and platinum profiles. Connection admission control
enables congestion control in the network, which facilitates
the achievement of individual QoS guarantee in the platinum
and gold profiles. Achievement of high network efficiency is
possible due to the incentives that come with the lowering
of tariffs for services in the platinum profile during periods
when the network has abundant resources. The characteristics
842
R.tatWVwly opvslndTlypic-al rqfouc avaILlabfilty
(a) Overprovisioned network (b) Typical resource availability
lr r-.rusc. allocationprCoS DMrn ."esuelo-atio per C-S
SW h.ring) (with SW sh-rbng)
o" a'o __ v
.. B.
PC X2=-
(c) No resource sharing (d) With resource sharing
Fig. 4. Network performance characteristics for the classes of service
DWe..tISW n-Otn pw p.rfUM
W
(a) No bandwidth sharing
DlffW.,,ti, SW *I9o n porpro(-h SW hs.ig)
(b) With bandwidth sharing
Fig. 5. Network performance characteristics of the profiles in one CoS
of different profiles enhance flexibility in the system, since
users can select the profile that suits their needs, and encourage
social fairness. Flat-rate pricing for the silver profile introduces
economic faimess to users who cannot afford the higher charges
of the gold and platinum profiles. The traffic control strategy
works on short time frames, which reflects the characteristic
nature of congestion in communication networks.
VII. CONCLUSIONS AND RECOMMENDATIONS
The DQBCM scheme introduces of service usage profiles,
and formulates an architecture for using policy-based admis-
sion control and DiffServ to achieve the QoS and charging
requirements of communication networks, i.e. improving of
network efficiency, controlling congestion, providing social and
economic fairness among users and maximising profits for the
network operators and service providers.
The network operators can configure the systems so that the
actual change in tariffs for the platinum profile occurs at certain
interval of I - S, as opposed to a linear tariff change format.
This will reduce the rate at which new CDR sessions need to
be created.
The application scenarios for the DQBCM scheme include:
3G networks, NGN, Internet service providers and telecom
operators who enforce SLAs with their users. In practical
deployment, the media access gateway, descnrbed in section IV,
can be built in a small gadget like a conventional residential
gateway device.
REFERENCES
[1] J. Gozdecki, A_ Jajszcyk, and R. Stankiewicz, "Quality of Service
Terminology in IP Networks," IEEE Communications Magazine, vol. 41,
pp. 153-159, March 2003.
[21 T. T. Ahonen, rn-Profits Making Money from 3G Services. John Wiley
and Sons, 2002.
[3} M. Falkner, M. Devetsikiotis, and I. Lambadaris, "An Overview of Pricing
Concepts for Broadband IP Networks," IEEE Communication Surveys,
vol. 3, pp. 2-13, Second Quarter 2000.
[41 J. Suvanto, "Multimedia Messaging Service for GPRS and UMTS," Proc.
IEEE WCNC, pp. 1422-1426, September 1998.
[5] N. Bagiaei and R. Hunt, "Review of Quality of Service Performance
in Wireless LANs and 3G Multimedia Application Services," Computer
Communications, vol. 27, pp. 1684-1692, November 2004.
[6] 3GPP, "Technical Specification Group Core Network; End-to-end Quality
of Service (QoS) signalling flows (Release 6)," Tech. Spec. 3GPP TS
29.208 V6.3.0, Mar. 2005.
[71 3GPP, "Technical Specification Group Services abd System Aspects;,
Telecommnunication management; charging management; Call and event
data for the Packet Switched (PS) domain," Tech. Spec. 3GPP TS 32.015
Y3.12.0, Dec. 2003.
[8] S. Blake and et. a], "An Architecture for Differentiated Services," RFC
2475, Dec. 1998.
[9] T. Li, Y. Iraqi, and R Boutaba, "Pricing and Admission Control for QoS-
enabled Internet," Computer Networks, vol. 46, pp. 87- 110, Sep. 2004.
[10] 3GPP, "Vocabulary for 3GPP Specifications (Release 5)," Tech. Ref 3GPP
TR 21.905 V5.8.0, Sep 2003.
[111 M. Karsten and et. al, "Charging for Packet-Switched Network Communi-
cations - Motivation and Overview," Computer Communications, vol. 23,
pp. 290-302, February 2000.
[12] N. Blefari-Melazzi, D. D. Sorte, and G. Reali, "Usage-based Pricing Law
to Charge IP Network Services with Performance Guarantees," IEEE,
vol. 25, pp. 2652-2656, Apr. 2002.
[13] A. Odlyzko, "Paris Metro Pricing for the Internet," Proc. ACM Conference
on Electronic Commerce, pp. 140-147, 1999.
[14] D. J. Songhurst, Charging Communication Networks - From Theory to
Practice. Elsevier Science B. V., first ed., 1999.
[151 R. Blake, D. Clark, and S. Shenker, "Integrated Services in the Internet
Architecture: an Overview," IETF RFC 1633, June 1994.
[16] J. Diederich, L. Wolf, and M. Zitterbart, "A mobile differentiated services
QoS model," Computer communications, vol. 27, pp. 1106-1114, 2004.
[17] 3GPP, "Technical Specification Group Core Network; Policy Control
Over Go interface (Release 6)," Tech. Spec. 3GPP TS 29.207 V6.2.0,
Dec. 2004.
[1] 3GPP, "Technical Specification Group services and System As-
pects; Telecommunication management; Charging management; Packet
Switched (PS) domain charging (Release 6)," Tech. Spec. 3GPP TS 32.251
V6.2.0, Mar. 2005.
[19] 3GPP, "Technical Specification Group Services and System Aspects;
3GPP system to wireless Local Area Network (WLAN) interworking;
System description (Release 6)," Tech. Spec. 3GPP TS 23.234 V6.0.0,
Mar. 2004.
[20] 3GPP, "Technical Specification Group Services and System Aspects;
Quality of Service (QoS) concept and architecture (Release 6)," Tech.
Spec. 3GPP TS 23.107 V6.3.0, June 2005.
[21] M.-J. Lin, H. Luo, and L. F. Chang, "A Linux-based EGPRS real-time
test bed software for wireless QoS and Differentiated Service studies,"
IEEE-ICC, vol. 25, pp. 1039-1044, April 2002.
[221 B. Hubert and et. al, Linux Advanced routing and Traffic management
HOWTO, July 2002.
[231 R. Sandilands, "Network Traffic Generator." Intermet.
http //www. sourceforge. netfprojects/traffic.
843