Content uploaded by Malamati Louta
Author content
All content in this area was uploaded by Malamati Louta
Content may be subject to copyright.
Content uploaded by Panagiotis G. Sarigiannidis
Author content
All content in this area was uploaded by Panagiotis G. Sarigiannidis
Content may be subject to copyright.
Load Dependent Resource Allocation in
Cooperative Multiservice Wireless Networks:
Throughput and Delay Analysis
Thomas Lagkas
Department of Informatics and Telecommunications
Engineering
University of Western Macedonia
Kozani, Greece
tlagkas@ieee.org
Dimitrios G. Stratogiannis,
Georgios I. Tsiropoulos
School of Electrical and Computer Engineering
National Technical University of Athens
Athens, Greece
dstratog@mail.ntua.gr, gitsirop@mail.ntua.gr
Panagiotis Sarigiannidis, Malamati Louta
Department of Informatics and Telecommunications Engineering
University of Western Macedonia
Kozani, Greece
psarigiannidis@uowm.gr, louta@uowm.gr
Abstract--Cooperative wireless networks supporting multiple
services necessitate the application of a robust bandwidth
allocation policy to ensure Quality of Service (QoS) provision
to different applications. In this work, a load dependent
bandwidth allocation technique is presented considering traffic
priority and buffer load in the relay nodes of a cooperative
communication network. An analytical approach for
bandwidth sharing is provided along with a delay analysis,
verifying that the proposed scheme can efficiently provide
traffic differentiation, satisfying, also, the QoS requirements in
terms of bandwidth, packet transmission rate and delay. The
results obtained by the analysis are validated via simulations,
confirming the improved network performance in terms of
throughput and delay.
Buffer load; delay; MAC; QoS; Simulation; Cooperative
wireless network
I. INTRODUCTION
Cooperative communications are based on the
improvement of the performance of a wireless network
achieved by the cooperation among different nodes and the
adoption of spatial diversity [1, 2]. The cooperation among
nodes of a wireless communication system leads to improved
system capacity, optimal spectrum use and increase of
system reliability, since bit error rate and outage probabilities
are decreased [1, 3]. In a cooperative relay communication
system, each node transmits each own data and collaborates
with other relay nodes, accepting and forwarding
appropriately packets received towards a destination node.
Data packets arriving from other nodes are queued in the
relay buffers of each node. The overall system throughput
increases, while aggravating queuing delay at the relay nodes
gives rise to higher end-to-end transmission time [4]. Thus,
buffer traffic load in the cooperative nodes has a critical
impact on the network performance parameters such as
throughput, delay and jitter.
Data packets of different applications require distinct
treatment by the network according to their Quality of
Service (QoS) characteristics and their priority. For example,
critical emergency communications data calls should be
favoured during the resource allocation process, satisfying
their QoS requirements. In this perspective, multiple services
support in a cooperative wireless network should involve the
development of an efficient resource allocation scheme in
conjunction with a prioritization mechanism for handling
different traffic types. Appropriate techniques providing
access and efficient medium sharing are mandatory.
In recent literature, various Medium Access Control
(MAC) protocols for cooperative networks have been
presented. In [5], the CoopMac protocol has been introduced
based on a set of new characteristics applied on the data and
control plane of the well known IEEE 802.11 MAC protocol.
In [6], the relay enabled Distributed Coordination Function
(rDCF) has been presented, which is based on the
accessibility of the relay nodes to forward data packets of
other nodes. Furthermore, a new protocol called Cooperative
MAC (CMAC) has been proposed in [3]. CMAC exploits the
spatial diversity of cooperative communications through a
retransmission technique of partially correct frames that are
combined to reconstruct the initial one. The above protocols
are variants of the IEEE 802.11 MAC protocol, which has
certain performance limitations in terms of bandwidth
sharing when applied to cooperative networks.
In this paper, a load dependent bandwidth allocation
technique is presented, considering buffer load in the nodes
of a cooperative communication network that overcomes the
constraints of existing protocols. The presented technique is
based on traffic differentiation according to its priority and it
can be considered as part of a Medium Access Control
(MAC) protocol. The main objective of this technique is to
provide enhanced QoS support, taking into account various
parameters such as packet priorities, throughput and delay.
Moreover, an analytical approach is provided to examine the
impact of buffer load on bandwidth sharing along with a
packet delay analysis considering a cooperative
communications scenario.
The rest of the paper is organised as follows. In Section
II, the load dependent bandwidth allocation technique
considering traffic priorities is studied. The simulator
developed to validate the proposed technique is described in
Section III. In Section IV, the simulation results and the
evaluation of the proposed technique are discussed. Finally,
conclusions are drawn in Section V.
II. LOAD DEPENDENT BANDWIDTH ALLOCATION
SCHEME
A. Bandwidth Allocation Scheme
A cooperative wireless communication system is
considered where different nodes may either generate new
data or receive and forward relay data packets transmitted by
other nodes. Data traffic is categorized based on its priority
and every node has a different packet buffer for each traffic
priority level supported, as defined in IEEE 802.11e standard
[7]. The objective of the proposed technique is to prioritize
data packets according to the QoS requirements of the
applications supported, while network resources are assigned
to nodes proportionally to their load. Specifically, the
designed scheme allocates resources to the packet buffers of
each node proportionally to their load and traffic priority.
The total bandwidth allocated to a node is equal to the sum
of the resources allocated to its buffers.
Let N denote the number of buffers in a single node. The
traffic load status of ith buffer is described by the quantity
()
i
Qt, which depends on the buffer priority and its load at a
given time t. That is
() ()
i
p
ii
Qt z Lt=, (1)
where
{}
1,...,iN∈, 1z≥ is a preset priority factor given by
the ratio of the bandwidth allocated to a buffer divided by the
bandwidth assigned to a buffer with equal load having the
consecutive lower priority, *
i
p∈Z is the priority of ith
buffer and ()
i
Lt denotes the actual load of ith buffer at a
given time t.
The normalized portion of bandwidth allocated to each
buffer is denoted by ()
i
normQ t , given by the following
equation:
1
() () ()
N
ii k
k
normQ t Q t Q t
=
=∑. (2)
Each node can be characterized by the sum of ()
i
Qt values
of all buffers within the node. Resource allocation within
each node is based on ()
i
normQ t , while the bandwidth
allocated to ith buffer depends on the ()
i
Qt value.
B. Packet Transmission Rate Analysis
Based on the sharing technique presented above,
bandwidth is allocated to each buffer of every node taking
into account its load and traffic priority. As previously
mentioned, for a specific node, data packets are either
generated from the node under consideration or arrive from
other nodes to be forwarded to a destination node. Despite
the fact that Packet Generation Rate (PGR) and bandwidth
allocation might be given, as shown in (1) and (2),
respectively, the channel access probability cannot be
determined directly since the allocation is affected by the
traffic flow priority. Consequently, to provide the requested
QoS levels for the traffic flows supported, appropriate
modifications to the priority levels or the bandwidth sharing
policy are necessary.
Suppose that there are two buffers (N=2) corresponding
to different traffic flows with discrete priority levels. Given
that ()
i
ft,
{
}
1, 2i=, is the total number of bits transmitted
at time t by the buffer i, the respective Packet Transmission
Rate (PTR) is represented by its first derivative, ()
i
ft
′
. In the
presented analysis, it is assumed that there is a bandwidth
capacity constraint where the buffers PGR is greater than the
available bandwidth b, otherwise, the PTR of each buffer
will be equal to the PGR, which is the ideal case. In what
follows, the node buffers PTR, c, is obtained by
12
1
12
() ()
1
11
2
(1),( 2 ) 1
1
12
() () ( ())
()
()
() 11
() ()
ft ft b
p
pp
ft cftcbft
ft
zLt
cb cb
ft b
cc
zLt z Lt
′′
+=
′
′′
= ⎯⎯⎯⎯⎯→=− ⎯⎯→
′
′
=
⎯⎯⎯→=⎯⎯→
++
+
12
12
() ()
pp
zLt czLt=. (3)
Without loss of generality, it is assumed that the PGR of
each buffer remains constant in time, therefore its mean i
a
can be used, e.g., the packet generation process follows the
Poisson distribution and its rate PGR is equal to the expected
mean value λ. Let Gi(t) denote the total load expressed as the
number of bits that have arrived to the ith buffer during the
observation interval [0, ]t. Evidently, the first derivative of
Gi(t) is equal to i
a. Since i
a is constant, Gi(t) is a linear
function of t. Moreover, the first derivative of fi(t) is also
constant in time, therefore fi(t) is also a linear function of t.
In this course, c can be obtained as follows
12
12
12
() () ()
11 22
()
11 2 2
12
(3) ( () ()) ( () ())
( ()) ( ())
11
iii
ii
Gt f t Lt pp
Gt a pp
ppr r
zGtftcGtft
z at f tt cat f tt
cb cb
za ca b
cc
−= −
′=−
−=
⎯⎯⎯⎯⎯⎯→−=−
′′
⎯⎯⎯⎯→−=−
⎛⎞
⎛⎞⎛⎞
⎯⎯⎯⎯→− =−− ⎯⎯→
⎜⎟
⎜⎟⎜⎟
++
⎝⎠⎝⎠
⎝⎠
2
22 1 1
()0
rr r
ac a b az bz c az+−− + − =. (4)
Solving the equation (4), c can be calculated by the
following quadratic formula
2
21 21 12
2
()()4
2
rr rr r
abazbz abaz bz aaz
ca
−−− + ± −− + +
=, (5)
considering only the value of c such that 0c≥, since the
PTR cannot be negative.
Following a general concept, suppose that multiple
buffers are employed by each node. It is considered that
traffic flows are categorized and grouped based on their
priority to different priority levels that correspond to virtual
buffers. The analysis of the proposed load dependent
approach for resource allocation considering multiple virtual
buffers can become quite complex. In this case, the PTR of
each buffer can be calculated via solving a system of
equations that is derived accordingly. The notations for the
general case employed are given below
• V denotes the number of virtual buffers, where
VN≤,
• ()
r
ft
′ is the PTR of the rth virtual packet buffer,
{}
,rij=,
{}
, 1,...,ij V∈, which is equal to the
bandwidth allocated to it,
• ij
b denotes the aggregated bandwidth assigned to i
and j buffers with ij≠ and
• () ()
ij i j
cb ft f t
′′
⎡⎤
=
⎣⎦ is the PTRs ratio, determined
by (5).
In what follows, an indicative case of multiple buffers
has been selected to demonstrate the features of the load
dependent allocation technique. Assume that there are three
virtual buffers 3N= and the total available bandwidth
capacity is b. Consequently, the following system of
equations is formed
12 12
1
12
[]
() 1[]
cb b
ft cb
′=+, (6)
23 23
2
23
[]
() 1[]
cb b
ft cb
′=+, (7)
12 1 2
() ()bftft
′′
=+, (8)
23 2 3
() ()bftft
′′
=+, (9)
123
() () ()
f
tbft ft
′′′
=− − . (10)
By solving the above system the 1()ft
′, 2()ft
′, 3()ft
′
and the
aggregated bandwidth assigned to buffers 12
b and 23
b are
obtained.
Note that in the presented approach buffers are assumed
to have infinite capacity. However, a more realistic approach
where buffers of limited capacity are considered should also
be examined. The constraint of limited buffer capacity will
result into two cases. In the first one, the buffer reaches its
maximum capacity, since it is assigned less bandwidth than
the PGR, while in the second one the bandwidth assigned is
adequate to serve its PGR, therefore, its load remains
constant. To determine which of the two cases occurs the
following algorithm is employed:
Algorithm for finite capacity buffers
• Step A: The algorithm estimates the available
bandwidth i
W for ith buffer as follows
1
i
p
i
iN
k
k
zC
Wb
C
=
=∑
, (11)
where, i
C is its maximum capacity.
• Step B: If Wi ≥ ai, then the PTR of ith buffer is set
equal to ai and the algorithm proceeds to the next
step. Otherwise, the algorithm returns to Step A and
the next buffer is examined until all the buffers are
assigned with PTR equal to ai or none of the
remaining buffers satisfies the condition of Step B.
• Step C: The PTR of ith queue as obtained during
the previous steps is subtracted from the total
available bandwidth. The process is repeated for
the remaining buffers taking into account each time
the remaining unallocated bandwidth.
• Obviously, if at least one buffer gets less bandwidth
than the required by its PGR, then all the buffers
will get full. In this case, the PTR of each buffer is
determined proportionally to the i
W.
C. Delay Analysis
One of the most important network performance metrics
is the delay within the buffer of a cooperative node since it is
a critical QoS parameter for various applications. The
proposed bandwidth allocation technique ensures that the
delay in the relay node is compliant with the QoS
requirements of the traffic flow. Since the mean transmission
rate of each traffic flow can be determined, it is feasible to
estimate the average corresponding delay. According to
Little’s law [8], the average system queue size equals the
jobs’ arrival rate multiplied by the average waiting time. In
the case of a cooperative network, the average system queue
size is equivalent to the average buffer load, the job’s arrival
rate corresponds to the mean generation rate and the average
Figure 1. Convergence of the PTRs ratio provided by the simulation to the
ratio determined by the analysis.
waiting time is equal to the average delay (d). Hence, for
flow i it holds
00
'
0
11
() ( () ())
1()
( ()) .
2
ii ii
ii
dLtdtGtftdt
PGR PTR
at f t t dt
ττ
τ
ττ
τ
τ
==−=
−
=− =
∫∫
∫ (12)
III. MODEL VALIDATION AND ANALYSIS
To validate the proposed resource allocation technique an
appropriate simulator has been developed in C#. In the
performed simulations two priority levels have been
considered. The actual PTR of ith buffer and the total
number of bits transmitted considering all buffers are
denoted by Si and b, respectively. The number of bits
entering the first or the second buffer is denoted by 1a and
2a, respectively. The simulator operation is implemented
via consecutive allocation cycles as follows.
Simulation Loop for Bandwidth Allocation
do
{
L1 += a1;
L2 += a2;
Q1 = Math.Pow( z, p1 ) * L1;
Q2 = Math.Pow( z, p2 ) * L2;
normQ1 = Q1 / (Q1 + Q2);
normQ2 = Q2 / (Q1 + Q2);
S1 = normQ1 * b;
S2 = normQ2 * b;
if ( S1 > L1 )
{
S1 = L1;
S2 = b - S1;
if ( S2 > L2 )
S2 = L2;
}
else if ( S2 > L2 )
{
S2 = L2;
Figure 2. Two flows PTRs ratio versus priority levels ratio for various
PGRs ratios.
S1 = b - S2;
if ( S1 > L1 )
S1 = L1;
}
L1 -= S1;
L2 -= S2;
} while <TERMINATION CONDITION>
The accuracy of the analysis presented is validated by the
convergence of the simulation results to the analytical ones,
as demonstrated in Fig. 1 for the PTRs ratio. Illustrative
numerical examples for variable number of operation cycles
are presented using different values for the ratio 12aa and
priority levels ratio 12pp
zz. The values of the network
parameters employed are: 1000b=, 21000a=, 2z
=
and
24p
=
. Note that the continuous lines correspond to the
simulation results, while dots are assigned to the results
obtained by the analytical formulas provided in Section II.
As demonstrated in Fig. 1, the PTRs ratio obtained by
simulations has a slight divergence from the numerical
results, when a low number of simulation cycles is
performed. However, when the number of cycles increases,
an excellent agreement between both results is observed.
IV. NUMERICAL RESULTS AND DISCUSSION
Numerical results for two cooperative network scenarios
are presented in this section, employing the proposed
bandwidth allocation technique. The examined cooperative
network involves source nodes which communicate with an
out of range destination node through intermediate relay
nodes. A traffic flow of different priority and rate is
generated from each source node and their packets are
buffered in the respective relay nodes before their
transmission to the destination. The respective packet buffers
are considered to have infinite capacity, which is a common
assumption when analyzing related queuing schemes [9].
The presented evaluation scenario examines the resource
allocation among the relay nodes when they share a common
medium to communicate with the destination node.
Figure 3. Two flows average delay ratio versus PGRs ratio for various
priority levels ratios.
A. Two Relay Node Scenario
The first scenario includes two source nodes, two relay
nodes and a common destination node. In Fig. 2, the two
relay nodes PTRs ratio is plotted with respect to their priority
levels ratio, for different PGRs ratios. It is evident that
priority levels ratio correspond to also higher values of the
PTRs ratio. Specifically, the rate of this increment rises for
higher PGRs ratio values.
The results regarding packet delay are depicted in Fig. 3,
where the impact of the two nodes PGRs ratio on their
average delay ratio for various priority levels ratio values is
considered. The average delay was calculated based on the
delay analysis provided in Subsection II.C. From Fig. 3 it is
observed that the average delay ratio increases with the
PGRs ratio. However, higher values of the priority levels
ratio correspond to lower values of the delay ratio.
B. Three Relay Node Scenario
The second cooperative network scenario under
consideration includes three source nodes, three relay nodes
and one common destination node. The respective
performance results regarding three different traffic flows
with different priorities originating from different nodes are
depicted in Fig. 4 and Fig. 5. These results are obtained via
solving the equation set related to the multi-buffer analysis
presented in Subsection II.B. The solution of the
corresponding system of five equations was derived
employing the fsolve function of MATLAB, which is a
variant of the Powell trust-region dogleg method described in
[10].
In Fig. 4, the PTR for the three traffic flows is plotted as
a function of the second flow priority and the third flow
priority. The priority of the first flow is considered fixed at
value 4, while all flows are characterized by the same PGR
value. It should be clarified that traffic priorities assigned to
different flows are independent each another. However, in
Fig. 4 and Fig. 5 fixed priority combinations are considered
for sake of results comparison. In Fig. 4, the flow priority
increases with respect to the other flow priorities. Moreover,
greater flow priority values correspond to also greater PTR
values for the specific traffic flow. Specifically, from Fig. 4
Figure 4. Three flows PTR versus varying combination of flow 2 and
flow 3 priorities.
Figure 5. Three flows PTR versus varying combination of flow 2 and
flow 3 PGRs.
it can be observed that the dependence of PTR on traffic flow
priority is non-linear.
Fig. 5 depicts the three flows PTR values for different
PGR values and the same priority level. Following the same
concept regarding the presented results, the PGR of the first
flow is fixed at 1000, while a fixed combination of the
second and third flows PGRs is considered. It is observed
that PTR increases proportionally to PGR, as already
expected, when the traffic flows considered are of the same
priority level.
V. CONCLUSION
In this paper a load dependent bandwidth allocation
technique is proposed. The presented technique takes into
account the traffic QoS requirements, such as its priority
level and the PGRs. The results obtained demonstrate that
the proposed technique can provide traffic differentiation and
efficient resource allocation along with minimum delay.
Numerical and simulation results were presented to validate
the proposed model. The flexibility of the proposed
technique indicates than it can be easily adapted as a part of
MAC protocol for cooperative wireless networks providing
QoS guarantees.
REFERENCES
[1] [1 ]Y.-W. Hong, W. J. Huang, F.-H. Chiu and C.-C.J. Kuo,
“Cooperative communications in resource-constrained wireless
networks,” IEEE Signal Process Mag., vol. 24, no. 3, pp. 47-57, May
2007.
[2] A. Nosratinia and A. Hedayat, “Cooperative communication in
wireless networks,” IEEE Commun. Mag, vol. 42, no. 10, pp. 74-80,
Oct. 2004.
[3] S. Shankar, C.-T. Chou and M. Ghosh, “Cooperative communication
MAC (CMAC) – A new MAC protocol for next generation wireless
LANs,” in Proc. International Conference on Wireless Networks,
Communications and Mobile Computing, vol. 1, 2005, 1-6.
[4] G. Bansal, V. Sharma, N.B. Mehta and E. Altman, “Relay load
balancing in queued cooperative wireless networks with rateless
codes,” in Proc. ICC, 2010, pp. 1-6.
[5] P. Liu, Z. Tao, S. Narayannan, T. Korakis and S. Panwar,
“CoopMAC: A cooperative MAC protocol for wireless LANs,” IEEE
J. Sel. Areas Commun., vol. 25, no. 2, pp. 340 – 354, Feb. 2007.
[6] H. Zhu and G. Cao, “rDCF: A relay-enabled medium access control
protocol for wireless ad-hoc networks,” IEEE Trans on Mob.
Comput., vol. 5, no. 9, pp. 1201-1214, Sep. 2006.
[7] IEEE Std 802.11e-2005, “IEEE Standard for Information
Technology--Telecommunications and Information Exchange
Between Systems--LAN/MAN Specific Requirements--Part 11
Wireless Medium Access Control and Physical Layer specifications,
Amendment 8: Medium Access Control Quality of Service
Enhancements,” 2005.
[8] J. D. C. Little, “A proof for the queuing formula: L= λ w,”
Operations Research, vol. 9, no. 3, pp. 383-387, 1961.
[9] A. K. Sadek, K. J. R. Liu, and A. Ephremides, “Cooperative Multiple
Access for Wireless Networks: Protocols Design and Stability
Analysis,” in Proc. 40th Annual Conference on Information Sciences
and Systems, 2006, pp. 1224-1229.
[10] M. J. D. Powell, A Fortran Subroutine for Solving Systems of
Nonlinear Algebraic Equations, Numerical Methods for Nonlinear
Algebraic Equations, P. Rabinowitz, ed., Ch.7, 1970.