An Adaptive Bandwidth Reservation Scheme
in Multimedia Wireless Networks
Xiang Chen and Yuguang Fang
Department of Electrical and Computer Engineering
University of Florida, Gainesville, FL 32611
Abstract—Next generation wireless networks target to provide
quality of service (QoS) for multimedia applications. In this
paper, the system supports two QoS criteria, i.e., the system
should keep the handoff dropping probability always less than a
predefined QoS bound, while maintaining the relative priorities
of different traffic classes in terms of blocking probability. To
achieve this goal, a dynamic multiple-threshold bandwidth
reservation scheme is proposed, which is capable of granting
differential priorities to different traffic class and to new ad
handoff traffic for each class by dynamically adjusting
bandwidth reservation thresholds. Moreover, in times of network
congestion, a preventive measure by use of throttling new
connection acceptance is taken. Another contribution of this
paper is to generalize the concept of relative priority, hence
giving the network operator more flexibility to adjust admission
control policy by incorporating some dynamic factors such as
offered load. The elaborate simulation is conducted to verify the
performance of the scheme.
I. INTRODUCTION
With the increasing demands for mobile multimedia
services such as audio, video and data, next generation
wireless networks are expected to provide quality of service
(QoS) for such multimedia applications to users on the move.
Since the services have inherently different traffic
characteristics, their QoS requirements may differ in terms of
bandwidth, delay and connection dropping probabilities. It is
the networks’ responsibilities to fairly and efficiently allocate
network resources among different users in order to satisfy
such differentiated QoS requirements of each type of service
independent of the others.
In order to guarantee QoS requirements of these services
while accommodating rapidly growing population of mobile
users, efficient connection admission control (CAC) schemes
have to be used. In [1]-[3], the well-known Guard Channel
scheme and some of its variations were proposed to give
higher priority to handoff connections over new connections in
voice traffic, whose performance depends largely on the
choice of the number of guard channels, which is mainly
based on a priori knowledge of the traffic patterns. Some
dynamic bandwidth allocation schemes are investigated in [4]
[5]. However, in [4], the priority among different traffic
classes, either in handoff traffic or new traffic, was not
addressed while [5] did not take fairness into consideration.
[6] uses predefined call blocking probability profile to
maintain the relative priorities among different classes of
traffic, whose underlying assumption, however, may not hold
in fast changing networks.
In this paper, we propose a dynamic, multiple-threshold
bandwidth reservation scheme (DMTBR) for multimedia
mobile wireless systems. Three bandwidth reservation
thresholds, G
1
, G
2
, and G
3
(G
1
< G
2
< G
3
) are used to guarantee
QoS requirements. When the network is under heavy traffic
load, in order to guarantee QoS provisioning, further step by
throttling new connection acceptance is taken. In summary,
this scheme has the following features:
• It gives differential priorities to both new and handoff
connections with different types of services.
• It maintains the relative priorities and fairness among
traffic classes by taking into account both user QoS
profile and real traffic condition, which generalizes the
concept of relative priorities among traffic classes.
• It updates the reservation thresholds periodically, hence is
able to respond to the changing network conditions
quickly and effectively.
The rest of this paper is organized as follows. The next
section describes the traffic model considered. Our scheme,
including the target QoS criteria, is presented in Section III.
Then, the calculation issues involved in the scheme are
addressed in Section IV. In Section V, the scheme is verified
through simulations combined with result analysis. Finally,
this paper is concluded in Section VI.
II. TRAFFIC
MODEL
The system under consideration is a multimedia wireless
network with a cellular infrastructure, comprising a number of
cells. We assume the system uses fixed channel assignment
(FCA), which means each cell has a fixed amount of capacity.
No matter which multiple-access technology (FDMA, TDMA
or CDMA) is used, we could interpret system capacity in
terms of bandwidth. In this paper, we assume a single number,
“effective bandwidth” [7], is adequate for guaranteeing desired
QoS for any connection with certain traffic characteristics.
Hereafter, whenever we refer to the bandwidth of a
connection, we mean its effective bandwidth. We assume each
cell has C bandwidth units (BU). There are two classes of
incoming traffic: 1) Class I—real-time traffic and 2) Class
This work was supported in part by the U.S. National Science Foundation
under Faculty Early Career Development Award ANIR0093241 and under
grant ANI-0220287.
GLOBECOM 2003 - 2830 - 0-7803-7974-8/03/$17.00 © 2003 IEEE
II—non-real-time traffic. Typically, class I traffic includes
voice and video service while class II traffic is comprised of
data services like email, file transfer and web browsing. The
arrivals are Poisson processes, with respective arrival rates
λ
rt
and
λ
nrt
. The connection duration times of both connections
follow exponential distribution, with means 1/µ
rt
and 1/µ
nrt
.
Furthermore, we assume that the cell residence time
distributions of these two kinds of connections are also
exponentially distributed, with means 1/
γ
rt
and 1/
γ
nrt
. The
number of BUs required by each real-time and non-real-time
connection is BW
rt
and BW
nrt
, respectively. These assumptions
are appropriate and commonly used in literature. Since time
spent by a connection in a cell is the minimum of connection
duration time and cell resident time, following the
assumptions, we can easily obtain that, for these two type of
traffic, the distribution of the time spent by a connection in a
cell is also exponentially distributed, with the mean of d
rt
=
1/(µ
rt
+
γ
rt
) and d
nrt
= 1/(µ
nrt
+
γ
nrt
).
III. PROPOSED
SCHEME
A. QoS Criteria
The first criterion is about the handoff connection
dropping probability (CDP), i.e., the probability of a handoff
connection being dropped when a handoff has to be made due
to user roaming from one cell to another. In wireless networks,
we usually set an upper bound for this probability, like the
following:
nrtnrtd
rtrtd
QoSP
QoSP
≤
≤
,
,
(1)
where
P
d,rt
, QoS
rt
and P
d,nrt
, QoS
nrt
are CDP and the allowable
maximum dropping probability for real-time and non-real-time
traffic, respectively.
The second criterion is to maintain the relative priority
among different type of traffic in terms of new connection
blocking probability (CBP). Obviously, if there is no such
criterion, it is unfair to the traffic classes that require larger
bandwidth. To address this problem, we assume in each traffic
class’s profile, there exists a parameter named traffic priority
weight,
W, indicating which priority level the traffic class will
have. This parameter is set during the negotiation between the
user and the network operator, taking traffic characteristics
into account. A smaller weight means a higher priority. To
achieve a better fairness in CBPs among all traffic classes, the
network may keep the CBPs satisfying the following equation:
nrt
rt
nrtb
rtb
W
W
P
P
=
,
,
(2)
where
P
b,rt
, W
rt
and P
b,nrt
, W
nrt
are CBP and the predefined traffic
priority weight for the two traffic classes, respectively.
Generally speaking, many factors play a role in
determining the CBP. Each traffic class’s actual CBP depends
on system capacity, offered traffic load of the traffic class, the
priority of the traffic class, the admission policy adopted to
fulfill the QoS criteria related to handoff traffic, the action the
network may take in times of congestion, and so on. While
some factors like system capacity or pre-assigned traffic
priority could be static, offered load, actions taken to deal with
network congestion are dynamic. In this sense, the criterion in
Eq. (2) is static and not fair enough since it fails to reflect the
real time network situation. Therefore, we generalize the
concept of relative priority and propose a more general way to
maintain the relative priority among different traffic class,
using the following formula:
nrt
rt
nrtb
rtb
W
W
P
P
α
=
,
,
(3)
Compared with Eq. (2), we add one factor,
α, on the right-
hand side of Eq. (3).
α can be thought of as a function of some
of the dynamic factors described above, representing
network’s real traffic conditions or some procedures
responding to traffic changes or QoS status.
Since offered load is one of the commonly used measures
of network traffic load, as one way to make
α concrete, we let
α be a function of offered load per cell for each traffic class.
Offered load can be defined as the product of each traffic
class’s traffic arrival rate, call holding time and normal
bandwidth, i.e.,
rtn
rtnrtn
rtn
BW
OL
)(
)()(
)(
µ
λ
=
(4)
Replacing
α with the ratio of offered load of real-time and
non-real-time traffic, we obtain the following:
nrt
rt
nrt
rt
nrtb
rtb
W
W
OL
OL
P
P
×=
,
,
(5)
In this way, we take into account the offered load of each
traffic class. In other words, we maintain the relative priority
by keeping the ratio of the CBP of each traffic class equal to
the product of their corresponding ratio of traffic load and the
weight pre-defined. The advantage of this approach will be
shown in Section V. Hereafter, we term the scheme satisfying
QoS Eq. (1) and (2) DMTBR_A, and term the scheme
satisfying QoS Eq. (1) and (5) DMTBR_G.
B. Connection Admission Policy
Based on the thresholds G
1
, G
2
, and G
3
, the admission
policy, including the adoption of throttling new connections
acceptance in case of heavy congestion, is shown in Fig. 1.
Note that for new connections, the thresholds
G
2
and G
3
are
not fixed for either of the two traffic classes; instead, their
roles switch depending on the network’s instantaneous
situation. Parameter
prob
rt
(or prob
nrt
) denotes the probability
of throttling and function
rand() generates a random number
belonging to [0, 1).
switch can be considered a Boolean sign,
which indicates the roles of
G
2
and G
3
.
C. Cooperations Among Cells
In cellular networks, traffic in different cells has
correlation. Hence, it is necessary and more efficient to deal
with network congestion in a cooperative manner to prevent
this from happening through admission control. We adopt this
idea in this scheme to cope with the situation where the
network is undergoing heavy traffic load.
GLOBECOM 2003 - 2831 - 0-7803-7974-8/03/$17.00 © 2003 IEEE
For a period of time, each cell measures CBP and CDP,
i.e.,
P
b, rt
and P
d, rt
(or P
b, nrt
and P
d, nrt
). We will count the times of
increasing the reservation thresholds for handoff traffic. Once
one reservation threshold is consecutively increased for a
certain number of times, say three times, the cell is deemed
experiencing heavy handoff traffic. In this case, to reduce the
potential incoming handoff traffic hence keeping CDP below
the upper bound, the cell will inform all of its neighbors to
further throttle the acceptance of new connections of the same
traffic class as the handoff traffic class in the current cell. One
method to achieve this is to admit the new connection request
with a certain probability generated online, which is called the
probability of throttling new connections. Details on how to
generate the probability are given in Section IV.
IV. CALCULATION
ISSUES
A. Calculation of G
1
, G
2
, and G
3
The scheme requires accurately adjusting the values of
these three thresholds,
G
1
, G
2
, and G
3
, for every period of time.
We assume
G
1
are the number of BUs that needs to be
reserved to deal with handoff real-time connections that will
arrive in a period of
d from now to the future, where d is the
corresponding expectation of channel holding time for real-
time traffic. If during period
d, there are m real-time
connections in cell
i which will leave cell i due to completion
or handoff, and
n handoff real-time connections which will
enter cell
i. Therefore, a total of m + n events will happen in
cell
i during period d. Let s be a sequence of these m + n
events and
S(m, n) be the set of all possible sequences which
may take place in
d. Let Y(s) denote the maximum net change
in the number of BUs allocated to real-time connections in cell
i in d corresponding to each specific s. We set G
1
equal to the
expected value of
Y(s), which can be obtained as shown in [4].
∑
∈
=
),(
),(
)(
nmSs
1
nmS
sY
G
(6)
where |
S(m, n)| is the cardinality of S(m, n). According to the
traffic model described in Section II, we can easily obtain the
parameters in Eq. (6). Because of space limit, we do not give
the calculation details.
In a similar way, we can calculate
G
2
', which is the number
of BUs that needs to be reserved to deal with handoff non-
real-time connections that will arrive in a period of
d', the
corresponding expectation of channel holding time for non-
real-time traffic.
Once we get G
1
and G
2
', G
2
is obtained as followes:
'
GGG
212
+= (7)
Note that we grant higher priority to handoff real-time
traffic over non-real-time traffic by use of the calculation
order of
G
1
and G
2
as described above.
Before calculating
G
3
, we calculate G
3
', which could be
thought of as the reservation threshold that could be used
either to reserve bandwidth for new real-time traffic against
new non-real-time traffic, or to reserve bandwidth for new
non-real-time traffic against new real-time traffic, depending
on the instantaneous relative priority status for the traffic
classes. The initial value for
G
3
' could be set as BW
rt
or BW
nrt
.
Therefore,
G
3
can be estimated like the following:
'
,
,
'
323
nrt
rt
3
GGG
NRTforifBW
RTforifBW
G
+=
=
(8)
B.
Adaptation of Bandwidth Reservation Thresholds
The techniques we used to estimate G
1
, G
2
' and G
3
' only
serves to provide a good initial value. To meet the QoS criteria
in a dynamically changing network environment, further
adaptation of these thresholds is needed.
In Fig. 2, up_th
1
, down_th
1
(0< down_th
1
< up_th
1
< 1) are
the threshold factors indicating when the measured CDP is
above up_th
1
*QoS
rt
or below down_th
1
*QoS
rt
, the threshold
will increase or decrease. Once the threshold is consecutively
increased for a certain number of times, denoted by time_th,
the cell will inform all of its neighbors to do throttling as we
described before. Pow(up
1
, v_index) refers to the v_index
power of up
1
(
> 1), in which v_index is an integer. We also
notice that when the measured CDP exceeds up_th
1
*QoS
rt
, we
immediately boost v_index to zero if it was negative in
previous step. In this way, this scheme is always able to be
responsive enough to fulfill the QoS bound criterion. The
portion of how to adapt G
2
' is omitted since it is similar to that
of adapting G
1
.
To guarantee the second QoS criterion, G
2
and G
3
are used
to make Eq. (2) or (5) hold. There are three parameters,
namely, switch, percentage and adj_index. switch is defined as
before. Percentage refers to the deviation error the scheme may
tolerate and the second criterion is still considered being met.
For instance, if percentage is set to be 0.1, this means, as long
as the ratio of the right-hand side and the left-hand side of Eq.
if (the incoming handoff connection is real-time)
if (available BUs >= BW
rt
) accept;
else reject;
else // the incoming handoff connection is non-real-time
if (available BUs >= G
1
+ BW
nrt
) accept;
else reject;
if (switch = = true)
if (the incoming new connection is real-time)
if (available BUs >= G
2
+ BW
rt
&& rand( ) <= prob
rt
)
accept;
else reject;
else // the incoming new connection is non-real-time
if (available BUs >= G
3
+ BW
nrt
&& rand( ) <= prob
nrt
)
accept;
else reject;
else // (switch = = false)
if (the incoming new connection is real-time)
if (available BUs >= G
3
+ BW
rt
&& rand( ) <= prob
rt
)
accept;
else reject;
else // the incoming new connection is non-real-time
if (available BUs >= G
2
+ BW
nrt
&& rand( ) <= prob
nrt
)
accept;
else reject;
Figure 1. Admission policy
GLOBECOM 2003 - 2832 - 0-7803-7974-8/03/$17.00 © 2003 IEEE
(2) or (5) is within the range [0.9, 1.1], the equations hold and
the QoS criterion is met. The role of adj_index is very similar
to v_index. However, in the adaptation here, we change up
3
(
>
1), according the value of adj_index in a way that, the larger
the absolute value of adj_index, the faster the adaptation speed.
This ensures the adaptation of G
3
can promptly respond to the
change of the incoming traffic and (or) QoS status. Finally,
when adj_index is less than a threshold, adj_index_th, which
means G
3
' is nearly zero, the scheme will reverse the parameter
switch, letting G
3
reserved for the other traffic class instead of
the current traffic class it is for.
C.
Probability of Throttling
Each cell keeps a J×K non-negative integer array A for
each traffic class for its neighbors. J is the number of traffic
classes and K is the number of neighboring cells. For real-time
and non-real-time traffic considered in this paper, J is equal to
2. If the cell’s ith (i = 0, 1, … K-1) neighbor sends a message
to the cell to throttle or de-throttle a real-time traffic class,
then A[0][i] is incremented or decremented by 1. It is similar
for non-real-time traffic. When making admission decision for
an incoming new connection request, the cell will use the
following equation to generate the probability of throttling
new connections:
])][[max(
])][[max(
i1A
nrt
i0A
rt
bprob
bprob
=
=
(9)
where b is a real number less than and close to 1, say 0.9.
V.
SIMULATION RESULTS AND ANALYSIS
In this section, we present the performance of our
proposed scheme through simulation carried out with OPNET
Modeler 8.0. The simulation model is a wrap-around model
[8]. The total number of BUs in each cell is 50. The number of
BUs each real-time or non-real-time connection needs is BW
rt
= 1 or BW
nrt
= 4. The real-time connection may be voice calls
and the non-real-time connection may represent file transfer or
web browsing. For real-time traffic, the mean duration 1/µ
rt
=
300 seconds and the mean cell residence time 1/
γ
rt
= 150
seconds. For non-real-time traffic, 1/µ
nrt
=1500 seconds and 1/
γ
nrt
= 750 seconds. On average, each connection will handoff
once during its lifetime. A handoff request will randomly
choose a destination from the six neighboring cells. We
assume that 25% of traffic is real-time traffic, and 75% of the
traffic is non-real-time traffic. New connections arrive
according to a Poisson process. According to the assumption,
86.96% of the new connection arrivals are real-time, and the
rest are non-real-time. For both DMTBR_A and DMTBR_G,
QoS
rt
= 0.01 and QoS
nrt
= 0.05. The ratio W
rt
/W
nrt
is equal to 1,
with the deviation error 10%.
Fig. 3 and 4 show CBP and CDP for both traffic classes, as
a function of average new connection arrival rate for both
DMTBR_A and DMTBR_G. Through calculation, we know
that arrival rate 0.1 connection/sec corresponds to about 110
Erlangs, which are 220% of the full load. In fig. 3, as
expected, we can see that, for DMTBR_A, the CBP for the
two classes are almost equal to each other. For DMTBR_G,
since the ratio of offered load of each traffic class is taken into
consideration, which is equal to 1:3, the ratio of CBP for the
two traffic classes is also about 1:3. This is consistent with Eq.
(2) or (5). Through direct calculation, we find out that as an
average, the CBP for real-time traffic is reduced 58.23% in
DMTBR_G compared to that in DMTBR_A, while the CBP
for non-real-time traffic is only increased 9.89% compared to
that in DMTBR_A. In fig. 4, both schemes successfully keep
the CDP of both traffic classes under the predefined QoS
bounds as expected, even when the network is experiencing
heavy traffic. Also, there is no big difference in these two
schemes in terms of CDP.
Next, the performance is investigated in terms of
throughput. The system throughput is defined as followes:
STNUMCELLC
cellainiconneachbyspenttimeBW
TP
i
i
*_*
.
∑
∗
=
(10)
where C, as mentioned before, is the total number of BUs
available in each cell, CELL_NUM is the total number of cells
in the entire network and ST is the total simulation time. It can
be observed that both schemes successfully achieve a stable
system throughput even under heavy traffic situation, as
//Assuming the initial values of G
1
, G
2
', G
3
' are already obtained.
time
1
= 0, time
2
= 0;
v_index = 0, d_index = 0;
switch = true;
if (P
d, rt
>= up_th
1
*QoS
rt
) {
if (v_index < 0) v_index = 0;
else v_index++;
G
1
= G
1
*pow(up
1
, v_index);
time
1
++;
if ((time
1
% time_th) = 0)
{asking neighboring cells to throttle; time
1
= 0;}
}
else if (P
d, rt
< down_th
1
*QoS
rt
) {
v_index--;
G
1
= G
1
*pow(up
1
, v_index);
time
2
++;
if ((time
2
% time_th) = 0)
{ask neighbors to de-throttle; time
2
= 0; }
}
G
2
= G
1
+ G
2
'; // Adaptation of G
2
' is omitted.
if (switch = = true){
if (P
b, rt
/ P
d, nrt
>= W
rt
/W
nrt
* [OL
rt
/OL
nrt
]* (1+percentage))
adj_index++;
else if (P
b, rt
/ P
d, nrt
<= W
rt
/W
nrt
* [OL
rt
/OL
nrt
]* * (1-percentage))
adj_index--;
}
else{
if (P
b, rt
/ P
d, nrt
>= W
rt
/W
nrt
* [OL
rt
/OL
nrt
]* (1+percentage))
adj_index--;
else if (P
b, rt
/ P
d, nrt
<= W
rt
/W
nrt
* [OL
rt
/OL
nrt
]* (1-percentage))
adj_index++;
}
G
3
' = G
3
'*pow(up
3
, adj_index);
if (adj_index < adj_index_th) {
adj_index = 0;
switch = ! switch;
}
G
3
= G
2
+ G
3
';
Figure 2. Reservation thresholds adaptation
GLOBECOM 2003 - 2833 - 0-7803-7974-8/03/$17.00 © 2003 IEEE
shown in fig. 5. The network throughput keeps increasing as
the offered load increases, showing very little difference from
each other. Combining the observation in fig 3 and 4, we see
clearly the advantage of DMTBR_G over DMTBR_A, i.e., the
benefit gained by generalizing the concept of relative priority.
The network does not lose anything (in terms of network
throughput); however, the user satisfaction for real-time new
traffic (in terms of CBP) is significantly increased while the
user satisfaction for non-real-time new traffic is only slightly
affected. Meanwhile, the user satisfaction for handoff traffic
(in terms of CDP) is well maintained for both schemes.
Finally, we consider the detailed throttling operations in
each cell. Fig. 6 shows the throttling probabilities for both
types of traffic, starting from the beginning of a simulation run
for arrival rate = 0.1 in cell 0 and cell 36. Cell 0 is in the center
and cell 36 is located in the edge in the simulation model. As
time passes, the throttling probabilities for real-time traffic are
almost 1, which means the neighboring cells of cell 0 or 36
rarely throttle the new connection acceptance. This is
consistent with Fig. 4, where the CDP for real-time traffic is
well kept below the predefined QoS bounds, indicating there is
no need to reduce the new connection admission for fulfilling
the first QoS criterion. For non-real-time traffic, as time passes,
the throttling probabilities first drop, then fluctuate around a
certain value after the network enters into a steady state. Thus,
we know that the cells keep the first QoS criterion for non-real-
time traffic with the help of cooperative neighbors, which
reduce the admission probability for new connections due to
non-real-time traffic when necessary.
VI.
CONCLUSIONS
In this paper, a dynamic multiple-threshold bandwidth
reservation (DMTBR) scheme is proposed to guarantee QoS
provisioning in wireless multimedia networks. By dynamically
updating the bandwidth thresholds according to network
traffic situation and QoS criteria, this scheme works well to
provide QoS guarantee and efficiently use network resource,
as shown in the simulation. Some of the benefits acquired by
generalizing the concept of relative priority are also shown.
Since the proposed scheme involves slight changes to the
architecture of the current wireless network, it can be easily
adopted by 3G and beyond wireless systems.
REFERENCES
[1] D. Hong and S. S. Rappaport, “Traffic model and performance analysis
for cellular mobile radio telephone systems with prioritized and
nonprioritized handoff procedures,” IEEE Trans. Veh. Technol., vol.
VT-35, no.5, pp. 77-92, Aug. 1986
[2] R. Guerin, “Queueing-blocking system with two arrival streams and
guard channels,” IEEE Trans. Commun., vol. 36, pp. 153-163, Feb.
1988
[3] C.-J. Chang, T.-T. Su and Y.-Y. Chiang, “Analysis of a cutoff priority
cellular radio system with finite queueing and reneging/dropping,”
IEEE/ACM Trans. Networking, vol. 2, no. 2, pp. 166-175, Apr. 1994
[4] P. Ramanathan, K. M. Sivalingam, P. Agrawal and S. Kishore,
“Dynamic resource allocation schemes during handoff for mobile
multimedia wireless networks,” IEEE J. Select. Areas Commun., vol. 17,
no. 7, pp. 1270-1283, Jul. 1999
[5] C. Oliver, J. B. Kim and T. Suda, “An adaptive bandwidth reservation
scheme for high-speed multimedia wireless networks,” IEEE J. Select.
Areas Commun., vol. 16, no. 6, pp. 858-874, Aug. 1998
[6] B. M. Epstein and M. Schwartz, “Predictive QoS-based admission
control for multiclass traffic in cellular wireless networks,” IEEE J.
Select. Areas Commun., vol. 18, no. 3, pp. 523-534, Mar. 2000
[7] A. I. Elwalid and D. Mitra, “Effective bandwidth of general Markovian
traffic sources and admission control of high-speed networks,”
IEEE/ACM Trans. Networking, vol. 2, no. 2, pp. 166-175, Apr. 1994
[8] I. Chlamtac, Y. Fang and H. Zeng, “Call blocking analysis for PCS
networks under general cell residence time,” in Proc. IEEE WCNC,
Sept. 1999
0 0.02 0.04 0.06 0.08 0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Arrival rate
Prob
ability (P
b
)
DMTBR_A RT
DMTBR_A NRT
DMTBR_G RT
DMTBR_G NRT
Figure 3. CBP vs. Arrival rate
0 0.02 0.04 0.06 0.08 0.1
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
Arrival rate
Prob
ability (Pd)
DMTBR_A RT
DMTBR_A NRT
DMTBR_G RT
DMTBR_G NRT
Figure 4. CDP vs. Arrival rate
0 0.02 0.04 0.06 0.08 0.1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Arrival rate
Thro
ughput
DMTBR_A
DMTBR_G
Figure 5. System throughput vs. Arrival rate
0 2000 4000 6000 8000 10000 12000
0.4
0.5
0.6
0.7
0.8
0.9
1
Time(sec )
Thro
ttling prob
ability
cell 0: RT
cell 36: RT
cell 0: NRT
cell 36: NRT
Figure 6. Throttling probability
GLOBECOM 2003 - 2834 - 0-7803-7974-8/03/$17.00 © 2003 IEEE
