Más contenido relacionado La actualidad más candente (19) Similar a On Demand Bandwidth Reservation for Real- Time Traffic in Cellular IP Network Using Evolutionary Techniques (20) On Demand Bandwidth Reservation for Real- Time Traffic in Cellular IP Network Using Evolutionary Techniques1. ACEEE International Journal on Network Security, Vol 1, No. 2, July 2010
On Demand Bandwidth Reservation for Real-
Time Traffic in Cellular IP Network Using
Evolutionary Techniques
M. Anbar, D.P. Vidyarthi
School of Computer and Systems Sciences, Jawaharlal Nehru University, New Delhi, India
Email: mohamm19_scs@mail.jnu.ac.in, dpv@mail.jnu.ac.in
Abstract— As real-time traffic requires more attention, it evolutionary algorithm can be summarized as follows
is given priority over non-real-time traffic in Cellular IP [2].
networks. Bandwidth reservation is often applied to serve Generate a population of individuals
such traffic in order to achieve better Quality of Service Repeat {
(QoS). Evolutionary Algorithms are quite useful in Test the individuals according to a fitness function
solving optimization problems of such nature. This paper Select individuals to reproduce
employs Genetic Algorithm (GA) for bandwidth Produce new variations of selected individuals
management in Cellular IP network. It compares the Replace old individuals with old ones
performance of the model with another model used for }
optimizing Connection Dropping Probability (CDP) using Until satisfied
Particle Swarm Optimization (PSO). Both models, GA
based and PSO based, try to minimize the Connection There are many Evolutionary Algorithms e.g.
Dropping Probability for real-time users in the network
Particle Swarm Optimization (PSO), Genetic
by searching the free available bandwidth in the user’s
cell or in the neighbor cells and assigning it to the real-
Algorithms (GA), Ant Colony Optimization (ACO) etc.
time users. Alternatively, if the free bandwidth is not Evolutionary Algorithms often offer well
available, the model borrows the bandwidth from non- approximating solutions to all types of problems. The
real time-users and gives it to the real-time users. proposed work uses Genetic Algorithms (GA) for
Experimental results evaluate the performance of the GA bandwidth management in Cellular IP networks and
based model. The comparative study between both the compares the performance of the model with the
models indicates that GA based model has an edge over performance of the PSO based model, proposed earlier
the PSO based one. [3], with the same objective.
Index Terms—Genetic Algorithm, Cellular IP networks,
Many algorithms for bandwidth management have
Quality of Service, Connection Completion Probability,
Bandwidth Reservation, Particle Swarm Optimization. been proposed in the literature. Adaptive Resource
Reservation schemes and bandwidth reservation using
I. INTRODUCTION Support Vector Machine and Particle Swarm
Optimization have been proposed in [4]. This model is
Wireless communications are considered to be the proposed in Cellular networks to avoid the unwillingly
most important development in communication forced termination and waste of limited bandwidth. The
technology. Services are getting better through scheme using both Support vector machine and Particle
generations; though in return, the problems involved Swarm Optimization was applied when handoff traffic
are also becoming more complicated. Further, is heavy. Probabilistic Resource Estimation and Semi-
providing good Quality of Service (QoS) for users in reservation scheme for bandwidth management has
wireless networks imposes more problems. Bandwidth been studied in [5]. In this model, the probability of
is the most important parameter in wireless networks, real usage of resources by the Mobile Host is
especially in Cellular IP networks, that affects the QoS considered using a probabilistic resource estimation
[1]. Bandwidth management falls in NP class of the and semi-reservation scheme. This scheme can improve
problems and thus soft computing methods can be connection blocking and connection dropping
applied to find a sub-optimal solution for call drop probabilities. Another approach that uses bandwidth
minimization by managing the bandwidth properly. management is Dynamic Grouping Bandwidth
Evolutionary Algorithms (EAs) are used for solving Reservation scheme for multimedia wireless networks
optimization problems that creates a big search space. which is based on probabilistic resource estimation [6].
These algorithms maintain a population of individuals According to this scheme, when the Mobile Host (MH)
(usually randomly generated initially) that evolves requests a new connection flow or it handoffs to a new
according to the rules of selection, cross-over, mutation cell, it provides some important information e.g. the
etc. All individuals are evaluated against a fitness estimated switching time and the estimated staying
function. The fittest individuals are more likely to be time etc. The main concern of this model is multimedia
selected for reproduction in the next generation. An traffic and QoS guarantee for this type of traffic;
38
© 2010 ACEEE
DOI: 01.ijns.01.02.08
2. ACEEE International Journal on Network Security, Vol 1, No. 2, July 2010
therefore the model establishes many reservation time problem (e.g. bandwidth reservation problem) can be
sections called groups according to the mobility transformed to the function optimization problem. As
information of Mobile Hosts. This model is used to an algorithm, the main strength of the PSO is its fast
reduce the connection blocking and connection convergence. Due to its well organized logic and
dropping rates. In [7] a framework for bandwidth procedures the optimal solution for a specific problem
management in ATM networks unified with traffic can be attained very fast. PSO shares many common
control has been proposed. The bandwidth required by points with the GA. Both algorithms start with a group
connections carried in each output port of the ATM of randomly generated populations and both have
switches is estimated. The estimation process in this fitness function to evaluate the population. Also, both
scheme takes into account both the traffic source update the population (search space) and search for the
declaration and the connection measurements at the optimal solution with random techniques [11].
output ports. In [8] two bandwidth strategies and PSO model is a swarm of individuals called
reservation scheme with fuzzy controller for real-time particles. Particles are initialized with the random
services has been discussed. The purpose of this model solutions. These particles move through many
is to support multiple types of service with different iterations to search a new and better solution for the
QoS requirements in heterogeneous wireless networks. problem. Each particle is represented by the two
The model presented in [9] allows a packet transfer in factors; one the position, where each particle has a
the switch and admits packets depending on the switch specific position and at the beginning initialized by the
and network occupancy. Packets are transferred if the initial position (x) and the other factor is the velocity
required bandwidth is smaller than the bandwidth (v), where each particle moves in the space according
currently available. Otherwise the packets are stored in to this velocity. During the iteration time (t), the
a buffer. In [10] an admission–level bandwidth particles update their position, and velocity [11].
management scheme consisting of Call Admission PSO simulates the behavior of the bird flocking.
Control (CAC) and dynamic pricing is proposed. The Consider the following scenario. A flock of birds is
main aim of this proposed scheme is providing randomly flying searching for the food in an area and
monetary incentives to users to use the wireless also there is only one piece of food in the area being
resources efficiently and rationally. searched. All the birds in the flock have learned that
The proposed model, in this work, provides better there is food in this area but none of them know where
QoS by on demand bandwidth reservation using the food is. The best strategy to locate the food is to
Genetic Algorithm in Cellular IP network. On-demand follow the nearest bird to the food [12].
refers to a service which addresses the cell’s need for In PSO algorithm there are two types of best values:
instant and immediate use. The performance of the one is Pbest which is the best position for each particle
proposed model has been evaluated by conducting the in the swarm and must be updated depending on the
experimental studies and its comparison with another fitness value for each particle. The second best value is
model. Gbest which is the global best value for the swarm in
The rest of the paper is organized as follows. In the general. This value must be checked, and is exchanged
next section, Particle Swarm Optimization (PSO) and by the best Pbest if the Pbest in this iteration is better
Genetic Algorithm (GA) have been briefly discussed. than Gbest for the last iteration.
Section 3 elaborates the proposed model. A The pseudo-code of the PSO algorithm is as follows.
comparative study, through experiment, has been done PSO ( )
between GA-based model and PSO-based model in {Initialize the swarm by giving initial and random
section 4. Section 5 contains some observations about values to each particle.
the experimental results. For each particle do
{Calculate the fitness function
II. PARTICLE SWARM OPTIMIZATION If the value of the fitness function for each
AND GENETIC ALGORITHMS particle at the current position is better than
the fitness value at pbest then, set the current
As the paper compares the two models, the tools
value as the new Pbest.
used in both of them have been briefed here.
A. Particle Swarm Optimization Choose the particle with the best fitness value
Swarm intelligence is an intelligent paradigm based of all the particles as the Gbest.
on the behavior of the social insects such as bird flocks,
fish school, ant colony etc. in which individual species Update the velocity of each particle as
change its position and velocity depending on its
k +1 k k k k k
neighbor.
Particle Swarm Optimization (PSO) is based on the V j = w.V j + c1.r1.( Pbest j − X j ) + c2.r 2.(Gbest − X j )
swarm intelligence. It is a population based tool used to Update the position of each particle as
find a solution to some optimization problems. The
39
© 2010 ACEEE
DOI: 01.ijns.01.02.08
3. ACEEE International Journal on Network Security, Vol 1, No. 2, July 2010
Selection: Through selection operation, good produced
X j = X j + V j.∆ t
k +1 k k
offspring (solutions) are selected depending on their
} fitness value for producing more offspring.
Until the solution converges. Cross-over: In this operation parts of two
} chromosomes are swapped to produce new offspring.
1In the pseudo code above, Cross-over can be either one site cross-over or multiple
k sites cross-over. Cross-over operation is done
Vj is the velocity of particle (j) in iteration (k).
according to a cross-over probability.
Pbest is the best achieved solution for each individual so far. Mutation: In mutation the parent chromosome is
Gbest is the global best value for the swarm.
changed by mutating one gene or more for yielding a
k
X is the current position of particle (j) in iteration (k). new offspring [15]. Mutation operation is done
j
according to mutation probability which is often low.
w is inertia weight and is varied from 0.9 till 0.4.
r1, r2 are random numbers between 0 and 1.
c1, c2 are acceleration factors that determine the relative III. PROPOSED MODEL
pull for each particle toward Pbest and Gbest The model, proposed in this work, applies GA for
and usually c1, c2 = 2. the optimization of the Connection Dropping
∆t is the time step and usually 1. Probability (CDP). The model has been described as
B. Genetic Algorithms follows. When a cell in a Cellular IP network requests
bandwidth for completing a real-time traffic
Genetic Algorithm is computerized search and transaction, the base station performs a GA based
optimization algorithm based on the mechanics of processing to reserve the available free bandwidth in all
natural genetics and natural selection. In general it is the cells. If it fails in doing so, the cell looks for the
not good to navigate through potentially huge search bandwidth assigned to the non-real time users. If it is
space for optimal solutions. It may incur huge amount not possible in any of the way as above the call is
of time. GA is a technique which can be applied in dropped. Other details are as below.
many cases to produce sub-optimal results during
reasonable amount of time. GA has many good features A. Assumptions
such as broad applicability, ease of use, and global Following assumptions have been made in the model.
perspective; therefore GA has been applied to various • A Cellular IP network of 50 cells is considered.
search and optimization problems in the recent past. • The cells are of hexagonal shape as shown in Fig.
Because of its population based approach, GA has also 3.
been extended to solve other search and optimization • Bandwidth is distributed among the cells
problems such as multi-objective and scheduling randomly and is divided into three parts: part I is
problems [13]. Population in GA consists of number of reserved for the users having real-time traffic,
individuals and each individual is considered as a part II is reserved for the users having non-real-
potential solution for the given problem. The individual time traffic and part III is free bandwidth in the
solution is also called chromosome [14] and consists of cell.
many genes, as shown in Fig. 1. The size of • Two typed of users (real-time and non-real-time)
chromosome depends on the type of the problem being are distributed randomly among the cells.
solved.
Data in the chromosome can be either in binary or B. Encoding Used
real as shown in Fig. 2(a) and 2(b). • Each solution (individual) is represented by a
A pseudo-code of the GA is as follows. chromosome.
GA ( ) • A chromosome is an array of length nine, as
{ Create a random population of any size; shown in Fig.4. Data representation in the
Evaluate the fitness function for each chromosome is real. The genes in the
individual in the population; chromosome are as follows.
For number of generations
{ Gene (0) is number of real-time users.
Select parents for reproduction; Gene (1) is number of non-real-time users.
Perform crossover (); Gene (2) is number of real-time packets.
Perform mutation (); Gene (3) is number of non-real-time packets.
Evaluate population; Gene (4) is size of real-time session.
} Gene (5) is size of non-real-time session.
} Gene (6) is bandwidth assigned for real-time users.
Gene (7) is bandwidth assigned for non-real-time
users.
Some of the functions used in the GA are as follows. Gene (8) is free bandwidth in the cell
40
© 2010 ACEEE
DOI: 01.ijns.01.02.08
4. ACEEE International Journal on Network Security, Vol 1, No. 2, July 2010
Genes mutating one gene of the parents’ chromosomes.
Mutation probability is 0.4.
D. Fitness Function
Following symbols are used in the fitness function.
Figure 1. Chromosome structure AB r : Available bandwidth for real-time users.
S r : Size of real-time session.
1 1 1 0 1 0 0 1 0
P s : Packet size.
Figure 2(a). Binary representation of data in the chromosome N p : Number of packets.
N r : Number of real-time users.
2 50 24 100 300 500 210 10 2 T s : Time of a session.
0 0 P r : Packet generation rate.
CCP : Connection Completion Probability.
Figure 2(b). Real representation of data in the chromosome
CDP : Connection Dropping Probability.
The model tries to minimize the fitness function CDP
33 as shown below:
32 30
29
ABr
31
25
26
CCP = (1)
28 21 Sr
27 24 20
19
23
19
15
CDP = 1 − CCP (2)
22 18 14
16
17 12 9 Available bandwidth for real-time users in the cell is
16
11 8
the amount of bandwidth given to the cell when
performing bandwidth distribution module.
10 7 5
6 4
Size of real-time session is calculated as follows.
3 2
1 S r =P s∗N p∗N r (3)
Figure 3. Network model used in GA
Number of packets generated in a session is:
C. Modules Used
N p =P r∗T s (4)
1) Bandwidth distribution: this module randomly
distributes the bandwidth among all the cells in
the Cellular IP network and divides the bandwidth From (1), (2), (3), and (4) the fitness function is
into three parts: real-time bandwidth (for real-
time users), non-real-time bandwidth (for non- AB r
real-time users) and free bandwidth. CDP = 1− (5)
P s∗P r ∗T s∗N r
2) User distribution: module for random users’
distribution among the cells and is classified into
two types: real-time users and non-real-time E. Algorithm
users. 1) Input population size (50 cells).
3) Borrow: this module performs borrowing 2) Input total number of users in the network.
bandwidth in the same cell (by borrowing the free 3) Input total amount of bandwidth to be distributed
bandwidth in it) or borrowing bandwidth from in the network.
another cell (by borrowing free bandwidth or 4) Distribute the users among cells.
bandwidth assigned to the non-real-time users 5) Distribute the amount of bandwidth among cells.
from another cell). 6) Generate the initial population
4) Cross-over: the cross-over operation is performed 7) Calculate the fitness function for all generated
between two chromosomes (two arrays) that chromosomes using (5).
generate two offspring from them i.e. two arrays. 8) Perform borrow function for cells that have CDP
Cross-over used in this algorithm is a single site less than 0.5.
cross-over with probability one. 9) For number of generations repeat the steps from 10
5) Mutation: this operation is used to generate new until 15
offspring that may have better fitness values by 10) Perform crossover function.
41
© 2010 ACEEE
DOI: 01.ijns.01.02.08
5. ACEEE International Journal on Network Security, Vol 1, No. 2, July 2010
11) Perform mutation function. 3) Number of generated real-time packets: 5, 10, 15,
12) Check the relevancy of data in the new generated 20, 25, Available bandwidth: 40 Mbps, Number of
chromosomes. users in the cell: 4 real-time, 4 non-real-time,
13) Calculate the fitness functions for the new Packet generation rate: 30 packets/sec.
generated population using (5). Time of the session has been generated randomly.
14) Sort the new generated population according to the
best fitness value. Changing Connection Dropping Probability with
different number of real-time packets
15) Select the best chromosome as the chromosome
with the best fitness value. 70
Connection Dropping
60
Probability(CDP%)
16) Store the results in an output file. 50
5 packets
10 packets
17) Consider the new population as the old population
40
15 packets
30
20 packets
for the next generation. 20
10
25 packets
18) Display the stored results in the output file. 0
1
4
7
10
13
16
19
Generation num ber
0 1 2 3 4 5 6 7 8
Figure 7. Effect of number of real-time packet generated on CDP
Figure 4. Chromosome structure used in the proposed model
4) Packet generation rates: 10, 20, 30, 40, 50
IV.SIMULATION EXPERIMENTS packet/sec, Available bandwidth: 40 Mbps,
Number of real-time users: 5 (same as in PSO
In this section, the performance of the proposed GA- based model).
based model is evaluated. The experiment has been
performed up to 20 generations with the given Time of the session has been generated randomly.
parameters. Changing Connection Dropping Probability with
1) Available bandwidth: 10, 20, 30, 40, 50 Mbps, changing packet generation rate
Number of users in the cell: 4 real-time, 4 non- 90
Connection Dropping
real-time; Packet generation rate: 30 packets/sec 80
Probability(CDP%)
70 Prate10
Time of the session has been generated randomly. 60
50
Prate20
Prate30
40
30 Prate40
20 Prate50
Changing Connection Dropping Probability with the
10
available bandwidth 0
13
19
11
15
17
1
3
5
7
9
100 Ge neration num ber
Connection Dropping
Probability (CDP%)
80 10 Mbps
20 Mbps
60
30 Mbps Figure 8. Effect of packet generation rate on CDP
40
40 Mbps
20
0
50 Mbps
5) Times of sessions in each cell: 2, 4, 6, 8, 10
minutes.
1
4
7
10
13
16
19
Generation num ber Available bandwidth: 40 Mbps, Packet generation
rate: 30 packet/sec, Number of real-time users: 5
Figure 5. Effect of available bandwidth on CDP (same as in PSO based model).
Time of the session has been generated randomly.
2) Number of real-time users in the cell: 5, 6, 7, 8, 9.
Available bandwidth: 40 Mbps, Packet generation Changing COnnection Dropping Probability with
different values of session's time
rate: 30 packets/sec.
Time of the session has been generated randomly. 80
Connection Dropping
70
Probability(CDP%)
2m
60
50 4m
Changing Connection Dropping Probability with 40 6m
30 8m
different number of real-time users in a cell 20
10 10 m
100 0
Connection Dropping
Probability (CDP%)
5 users
1
3
5
7
9
13
19
11
15
17
80
6 users Generation num ber
60
7 users
40
8 users
20 9 users Figure 9. Effect of session’s time on CDP
0
6) Comparison between GA based and PSO based
1
4
7
10
13
16
19
Generation num be r
model.
The proposed model has been compared with the
Figure 6. Effect of number of real-time users in the cell on CDP PSO based model for the following parameters and for
the same number of generations/iterations which is 20.
42
© 2010 ACEEE
DOI: 01.ijns.01.02.08
6. ACEEE International Journal on Network Security, Vol 1, No. 2, July 2010
i. Available bandwidth: 10, 20, 30, 40, 50 Mbps.
Number of users in the cell: 4 real-time, 4 non- v. Session times: 2, 4, 6, 8, 10 minutes, Available
real-time, Packet generation rate: 30 packet/sec. bandwidth: 40 Mbps, Packet generation rate: 30
Time of each session has been randomly packet/sec, Number of real-time users: 5 (same as
generated. in PSO based model).
Comparison be twe en PSO and GA for Comparison betwee n PSO and GA in sense of
available bandwidth session times
30 50
25 40
(CDP%)
20
(CDP%)
30 G.A
PSO
15
G.A 20 PSO
10
10
5
0
0
10 20 30 40 50 2 4 6 8 10
available bandw idth (Mb/sec) s e s sion tim e (m inute s )
Figure 10. CDP with changing the bandwidth Figure 14. CDP with changing the time of the session
ii. Number of real-time users in the cell: 5, 6, 7, 8, 9. V. OBSERVATIONS AND CONCLUSIONS
Packet generation rate: 30 packets/sec.
Available bandwidth: 40 Mbps. All experiments have been conducted in Cellular IP
Time of the session has been generated randomly. network that has many parameters affecting QoS such
as: number of real-time users in the network, available
Comparison betwe en PSO and GA for number
bandwidth, packet size, packet generation rate, time
of real-time users
taken for a complete session. The effect of each of the
30
25 mentioned parameters has been studied in the proposed
work. In all the experiments, packets of random sizes
20
(CDP%)
G.A
15
PSO
10
5
are generated with maximum size being 100 bytes and
0
5 6 7 8 9
the available bandwidth is taken care of as limited
number of real-time users
resource. The model is trying to optimize Connection
Dropping Probability (CDP) for real-time users using
Figure 11. CDP with changing number of real-time users Genetic Algorithm. The model also compares the
results obtained with a similar model using Particle
iii. Number of real-time packets 5, 10, 15, 20, 25
Swarm Optimization (PSO) algorithm.
packets, Available bandwidth: 40 Mbps,
Number of users in the cell: 4 real-time, 4 non-
It is observed that increasing the available bandwidth
real-time, Packet generation rate: 30 packets/sec.
leads to decrease in CDP for real-time traffic as is clear
Time of the session has been generated randomly.
from Fig. 5. Comparing GA based model with PSO
based model, in terms of this parameter, shows that
Comparison between PSO and GA in sense of
number of real-time packets both of them reduces the CDP when the available
30
25
bandwidth is more with the notice that GA is
performing better in reducing CDP as obvious from Fig
(CDP%)
20
G.A
15
10.
PSO
10
5
0
5 10 15 20 25
num ber of real-tim e packe ts
When number of real-time users increases, the
demand on bandwidth increases; therefore, CDP is
Figure 12. CDP with changing the number of real-time packets bigger every time there are more real-time users in a
cell as shown in Fig 6. From comparison with PSO
iv. Packet generation rate 10, 20, 30, 40, 50, based model, it is clear that CDP is going up when real-
Available bandwidth: 40 Mbps, Number of real- time users are increasing in number; though both GA
time users: 5 (same as in PSO based model). and PSO are controlling CDP below (0.5) with better
Time of the session has been generated randomly. values for CDP in case of GA. This observation is
derived from Fig 11.
Comparison between PSO and GA in sense of
Pa cke t Genera tion Rate
60 When real-time users generate bigger number of
packets, the consumed bandwidth is more and it results
50
40
(CDP%)
G.A
in bigger CDP with fixed amount of bandwidth. Both
30
PSO
20
10
0
10 20 30 40 50
the models (PSO-based and GA-based) are able to
Pack et Generation Rate (packe t/s ec) handle this problem easily but the GA based model is
performing better as shown in Fig 12.
Figure 13. CDP with changing the packet generation rate
43
© 2010 ACEEE
DOI: 01.ijns.01.02.08
7. ACEEE International Journal on Network Security, Vol 1, No. 2, July 2010
It is observed that when the rate of packet generation using particle swarm optimization”, Int. J. of Bus. Data
is bigger the required amount of bandwidth to complete Comm. and Netw. (IJBDCN), vol. 5, pp 53 - 65, 2009.
a call is also bigger as is clear from Fig 8. It has been [4] C.J. Huang, Y.T. Chuang, W. K. Lai, Y.H. Sun, and
C.T. Guan, “Adaptive resource reservation schemes for
observed that when real-time users generate more
proportional DiffServ enabled fourth-generation mobile
packets, PSO and GA both tolerate the increment in communications system”, Comp. Comm. J., vol. 30, pp.
packet generation rate. Though CDP increases in both 1613-1623, 2007.
the models, but it is still less than the value that drops a [5] G.S. Kuo, P.C. KO, and M. L. Kuo, “A probabilistic
connection. CDP values obtained from GA model are resource estimation and semi-reservation scheme for
less than those which have been obtained from PSO flow-oriented multimedia wireless networks”, IEEE
model as shown in Fig 13. Wire. Comm. and Netw. Conf., (WCNC 2000), vol. 3, pp.
1046 – 1051, 2000.
The effect of session’s time is not less important [6] J.Y. Chang, and H.L. Chen, “Dynamic-Grouping
bandwidth reservation scheme for multimedia wireless
than the effect of the other factors on CDP according to
networks”, IEEE J. of Selected Areas in Comm., vol. 21,
(5). When the time of a real-time session increases pp. 1566 – 1574, 2003.
there is a possibility for generating more packets and [7] Z. Dziong, M. Juda, and L.G. Mason, “A framework for
consuming more bandwidth. GA is performing better bandwidth management in ATM networks – aggregate
than PSO in controlling CDP with the increment in equivalent bandwidth estimation approach”, IEEE/ACM
session’s time as obvious from Fig. 14. Trans. on Netw. vol. 5, pp. 134 – 147, 1997.
[8] I.S. Hwang, B.J. Hwang, L.F. Ku, and P. M. Chang,
The discussion is concluded stating the reason “Adaptive bandwidth management and reservation
behind the better performance of the GA based model. scheme in heterogeneous wireless networks”, IEEE Int.
Conf. on Sensor Netw. Ubiquitous and Trustworthy
PSO model does not use crossover operation (i.e. there Computing, 2008. SUTC '08, pp. 338-342, 2008.
is no material exchange between particles) that makes [9] A. Hac, “Bandwidth management in the switch with
the particles same without change but they are various traffic Burstiness”, Third IEEE Conf. on
influenced by their own previous best positions and Telecomm. , pp. 343-347, 1991.
best positions in the neighborhood in the global [10] B. Al-Manthari, N. Nasser, N. A. Ali, and H. Hassanein,
population. In GA, there is a crossover operation (i.e. “Efficient bandwidth management in broadband wireless
there is exchange in the material between the access systems using CAC-based dynamic pricing”,
individuals in the population) that means there is a 33rd IEEE Conf. on Local Computer Netw., pp. 484-
491, 2008.
chance to generate new offspring with better
[11] N. Nedjah, and L. de M. Mourelle, Swarm Intelligent
specifications than the parents. GA model is better in Systems, Springer-Verlag Berlin Heidelberg, 2006, pp 3-
sense of values obtained in every generation but PSO 57.
model is better in sense of time taken for the [12] J.H. Seo, C.H. Im, C.G. Heo, J.K. Kim, H.K. Jung, C.G.
convergence. Lee, “Multimodal function optimization based on
particle swarm optimization”, IEEE Trans. on Magn. ,
REFERENCES vol 42, pp. 1095 – 1098, 2006
[13] D.E Goldberg, Genetic Algorithms in search,
[1] X. Yang, and G. Feng, “Optimizing admission control optimization, and Machine Learning, Upper Saddle
for multi-service wireless networks with bandwidth River, NJ: Pearson, 2005, pp. 1-25.
asymmetry between uplink and downlink”, IEEE Trans. [14] L.M.O. Khanbary, D.P. Vidyarthi, “A GA-based
Veh Tech., vol. 56, pp. 907 – 917, 2007. effective fault-tolerant model for channel allocation in
[2] L. C. Jain, V. Palade, and D. Srinivasan, Advances in mobile computing”, IEEE Trans. on Veh. Tech., vol. 57,
Evolutionary Computing for System Design, Springer- pp. 1823-1833, 2008.
Verlag Berlin Heidelberg, 2007, pp 1- 139. [15] Z. Michalewicz, Genetic Algorithms + Data Structures
[3] M.Anbar, and D.P.Vidyarthi, “On demand bandwidth = Evolutionary Programs, 3rd revised and extended Ed.,
reservation for real-time users in cellular IP network Springer, Charlotte, 1995, pp. 45-88.
44
© 2010 ACEEE
DOI: 01.ijns.01.02.08