Page 1
International Conference on Intelligent Information Systems and Management (IISM’2010), June10-12, 2010
Abstract──This paper presents generalized particle approach to
optimize the bandwidth allocation and QoS parameters in ATM
networks. The generalized particle approach algorithm (GPAA)
transforms the network bandwidth allocation problem into dynamics
and kinematics of numerous particles in a force-field. In ATM
networks optimization, the GPA deals with a variety of random and
emergent phenomena, like congestion, failure, and interaction. The
bandwidth allocated, success rate and the shortest path determined by
GPAA are compared with the same parameters determined using
Max-Min fairness (MMF) algorithm. The comparison among the
algorithms gives the effective performance analysis of ATM
networks.
I. INTRODUCTION
TM network is to support multiple classes of traffic (e.g.
video, audio, and data) with widely different
characteristics and Quality of Service (QoS) requirements.
The major challenges are guaranteeing the QoS required for
all the admitted users and dynamically allocating appropriate
resources to maximize the resource utilization. Many
algorithms and evaluation criteria for the bandwidth allocation
have been proposed are based on maximizing an aggregate
utility function of users, an average priority or average
reliability of traffics; minimizing the overhead of data
transmission and the highest call blocking Probability of all
source-destination pairs. In some duality models the network's
optimization problem can be solved by a primal Parallel
algorithm for allotted rates and a dual parallel algorithm for
shadow prices or congestion signals [1].
In this paper GPAA is used to determine bandwidth to be
allocated based on pricing and shortest path. MMF is used to
determine bandwidth to be allocated based on fairness and
pricing. The success rate, shortest path are also calculated
using both methods separately and compared.
II. GENERALIZED PARTICLE APPROACH
The GP model consists of numerous particles and forces,
with each particle having its own dynamic equations to
represent the network entities and force having its own time-
varying properties to represent various social interactions
among the network entities. Each particle in GP has four main
characteristics [2]:
M. Dhivya is with the Department of Electrical and Electronics
Engineering, Anna University, Coimbatore, India. e-mail:
[email protected] .
M. Sundarambal is with the Department of Electrical and Electronics
Engineering, Coimbatore Institute of Technology, Coimbatore, India.
L. Nithissh Anand is with the Department of Mechanical Engineering,
Anna University, Coimbatore, India
Digital Object Identifier No:IISM00620100001
1. Each particle in GP has an autonomous self-
driving force, to embody its own autonomy and the
personality of network entity.
2. The dynamic state of every particle in GP is a piecewise
linear function of its stimulus, to guarantee a stable
equilibrium state.
3. The stimulus of a particle in GP is related to its own
objective, utility and intention, and to realize the multiple
objective optimizations.
4. There is variety of interactive forces among particles,
including unilateral forces, to embody various
social interactions in networks.
The bandwidth allocation policy of a network link depends on
the payment policy of a network path. It changes dynamically
according to the situation of the demands and supplies of
bandwidths and influences the kinematics and dynamics of the
link particles and path particles in their force-fields. The link
particles move in resource force-field FR and the path particles
that require bandwidth assignment move in demand force-field
FD as shown in figure1. Normally a network path pays less
amount under the condition that its bandwidth requirement is
satisfied. It is embodied in the proposed model by the
corresponding path particle. Similarly to maximize the price
benefit a link tries to assign its own bandwidth to the path that
pays maximal price.
III. MODEL FOR BANDWIDTH ALLOCATION
3.1 EVOLUTION OF GP:
The mathematical model based on GP [1] for dynamic
bandwidth allocation that involves m links, n paths and p
channels is defined with following notations:
ri : The maximum bandwidth of link Ai.
d j
: The bandwidth that channel T j requires.
q j : The maximum payoff that channel T
j can
afford for its d j.
Ai : The ith physical link in the network.
Tk : The kth path of the jth channel T j.
a ik : The bandwidth that link Ai allots to path Tk.
P ik : The price per unit bandwidth that path Tk tries to
Pay link Ai.
Bandwidth Allocation in ATM Networks Using
Novel Particle Approach M. Dhivya, M. Sundarambal and L. Nithissh Anand
A
ISBN-978-1-4507-2041-0 © by CiiT 2010 Published by Coimbatore Institute of Information Technology 1
Page 2
International Conference on Intelligent Information Systems and Management (IISM’2010), June10-12, 2010
3.1.1Total Utility:
u k(t) is the distance from the current position of the path
particle T k to the upper boundary of the demand force-field FD
at time t, and let JD(t) be the total utility of all the path
particles in FD. uk(t) and JD(t) are defined respectively as stated
below;
m
u k(t) =δ 1 exp[- ∑ a ik(t) / p ik(t) ] (1) i=1
p n
J D (t) = α1 ∑ ∑ u k(t) (2) j=1 k=1
u i (t) is the distance from the current position of the link
particle Ai to the upper boundary of the resource Force-field
FR at time t, and let JR(t) be the total utility of all the link
particles in FR. ui(t) and JR(t)are defined respectively as stated
below;
p n
u i(t) =δ 2 exp [- ∑ ∑ p ik(t) / a ik(t) ] (3) j=1 k=1
m
J R (t) = α2 ∑ u i(t) (4)
i=1
δ 1, δ 2 >1,and 0< α1, α2 <1.
3.1.2 Potential Energy:
At time t, the potential energy functions, PD(t) and PR(t),
that are caused by the upward gravitational forces of the force-
fields, FD and FR, are defined respectively as stated below;
p n
PD(t) = ε 2 ln[- ∑ ∑ exp [( u k(t))
2 / 2 ε
2] - ε
2 ln(n) (5)
j=1 k=1
m
PR(t) = ε 2 ln ∑ exp [( ui (t))
2 / 2 ε
2] - ε
2 ln(m) (6)
i=1
At time t, the potential energy functions, QD(t) and QR(t),
that are caused by the interactive forces among the particles
in FD and FR are defined respectively as stated below;
p n
Q D (t)= β1 ∑ [∑ a ik(t) – d j(t)]
2 + E D(t)
j=1 k=1
p n m
+ ρ∑ [ ∑ ∑ a ik(t) p ik(t) – q j(t)]
2 (7)
j=1 k=1 i=1
m n p
Q R (t) = β2 ∑ [∑ ∑ a ik(t) – r j(t)]
2 + ER(t) (8)
i=1 k=1 j=1
0< ε<1, 0< β1 β2, ρ<1;
ED(t) and ER(t) are the potential energy functions that involve
other kinds of the interactive forces among the particles in
FD and FR, respectively.
3.1.3 Stability:
Dynamic equations for path particle T k and link particle Ai
are defined respectively as stated below;
du k (t)/dt =φ1(t)+φ2(t) (9)
φ1(t) = - u k(t) + γ1 v1 (t) (9a)
φ2(t) =-[ε1 +ε2 dJ D(t)/d u k(t) + ε3dPD(t)/du k(t)
+ ε4 dQD(t)/d u k(t)] *
m
∑ [d u k(t) / dp ik(t)]2 (9b)
i=1
(OR)
du i (t)/dt = ψ1(t)+ ψ 2(t) (10)
ψ 1(t) = - u i(t) + γ2 v2(t) (10 a)
ψ 2(t) = -[ λ 1 + λ 2 dJ R(t)/d u i(t)+ λ 3 dPR(t)/d ui (t)
+ λ 4 dQR(t)/d u k(t) ] *
n p ∑ ∑ [d u i (t) / da ik(t)]
2 (10 b)
k=1 j=1
v1(t) and v2(t) is a piecewise linear function.
γ1, γ2 >1; 0< λ 1, λ 2, λ 3 ,λ 4, ε1, ε2, ε3 ,ε4,<1
dp ik(t)/dt = - ε1[ d u k(t) / dp ik(t)] - ε2[dJ D(t)/dp ik(t)]
-ε3[dPD(t)/dp ik(t)] - ε4[dQD (t)/dp ik(t)]
(11)
da ik(t)/dt = - λ 1[ d u i(t) / da ik(t) - λ 2[dJ R(t)/da ik(t)]
-λ 3[dPR(t)/ da ik(t)] - λ 4[dQR (t)/da ik(t)]
(12)
3.2 GP ALGORITHM:
Based on the above mathematical model the algorithm is
written as given below.
Input:
Maximum bandwidth ri , required bandwidth d
j ,
Maximum payoff q j
Output:
1. Initialization:
Figure1: GPAA to optimize bandwidth allocation-
a) Demand force- field FD for path particles. b) Resource force-field FR for link particles.
ISBN-978-1-4507-2041-0 © by CiiT 2010 Published by Coimbatore Institute of Information Technology 2
Page 3
International Conference on Intelligent Information Systems and Management (IISM’2010), June10-12, 2010
At time t=0;
Initiate bandwidth a ik(t) and price p ik(t)
2. calculate utility, potential energy functions as per the
Equations (1) to (8).
3. If
du k (t)/dt =0;
(Or) du i (t)/dt =0;
hold for every particle ,finish with success;
Else goto step 4.
4. Calculate du k(t)/dt and update u k(t)
calculate dui (t)/dt and update ui(t)
calculate da ik(t)/dt ,
a ik(t) = a ik(t -1) + da ik(t)/dt
calculate dp ik(t)/dt,
p ik(t) = p ik(t -1)+ dp ik(t)/dt
goto step 2.
A network with 7 nodes and 11 edges as exhibited in
Figure 2: is considered with the set of node pairs:
V ={(v1,v7),(v2,v6),(v4,v6),(v1,v3),(v1,v5)}
Figure 2: A network with 7 nodes and 11 edges
The communication bandwidth bij requested between given
set of (v=vi,vj) network node pairs are:
B ={15,25,21,20,17}
The virtual path sets are:
T18 ={(e8,e9),(e1,e2,e3),(e1,e11,e9),(e1,e2,e12,e9)}
T27 = {(e2, e3,e4),(e11,e10,e5),(e11,e9,e4),
(e2, e12,e10,e5), (e2,e12,e9,e4)}
T47 ={(e10,e5),(e9,e4)}
T16 ={(e7,e6),(e8,e10),(e1,e11,e10)}
T26 ={(e11,e10),(e2,e12,e10)}
And the set of links with maximum allowable bandwidths are
{e1=10 ;e2=8; e3=3; e4=18; e5=19; e6=8; e7=12; e8=15;
e9=4; e10=6;e11=15;e12=20}
The minimal price, bandwidth requested, maximal price,
maximum bandwidths of link are given as input, to the GPA
algorithm. The QOS index required by every virtual path
connection (VPC) is generated randomly within 0.5 to 1.5.
IV. MAX-MIN FAIRNESS ALGORITHM
The Max-Min fairness is computed via an iterative
procedure. The algorithm classifies every virtual session as
either completely utilized or not[3].
The allocated rate for session P is denoted by rp and the
allocated flow on network is[4];
Fa= ∑ δp(a) r p
pεP
δp(a)=1 if a is on path P and 0 otherwise.
r p>=0, for all p ε Pk
Fa<=Ca, for all a ε A.
Ca is capacity of arc a and r is vector of allocated rates.
4.1 MMF ALGORITHM :
Initial Conditions:
K=1,where k is the iterative index.
Capacity Fa =0; rate r p=0; P1=P; A
1=A;
1. nk
=number of paths ; p ε Pk
with δp(a)=1
2. rate vector r- k = min(Ca-Fa(k-1))/n
k
3. r p = r p(k-1)+ r- k for p ε Pk
Else
r p = r p(k-1)
4. Fa = ∑ δp(a) r p
pεP
5. A(k+1) = {a|Ca-Fa > 0}
6. P(k+1) = {p| δp(a)=0,for all ¢ A(k+1)}
7. k = k+1
8. If P k is empty, then stop; else goto 1.
The above algorithm terminates and finds the max-min
fairness vector r, if it exists, within k steps.
Initially all sessions are unsaturated, and their status change
from unsaturation to saturation. A session is allocated a rate r p
equal to minimum of the link bandwidth on its path. Initially
bandwidth allocated to the link is one third of the minimum
bandwidth of link on its session P. The algorithm checks the
saturation condition and updates the bandwidth till the session
is saturated. The algorithm terminates if all the sessions are
saturated.
For the network in figure 2: the bandwidth requested,
minimal price (rp) and the allowable bandwidth of link (Ca)
are given as input to MMF algorithm.
V. PERFORMANCE ANALYSIS
The Generalized particle approach (GPAA) and Max-Min
fairness algorithm is implemented with MATLAB 7.0.1.
The solution obtained is given below.
12
2
1 8 4
5
3 6 7
10
8
8 19
18 6
14 4
20
15 3
ISBN-978-1-4507-2041-0 © by CiiT 2010 Published by Coimbatore Institute of Information Technology 3
Page 4
International Conference on Intelligent Information Systems and Management (IISM’2010), June10-12, 2010
TABLE 1: BANDWIDTH ALLOCATION
Path
path set
Bandwidth allocation
GPAA MMF
T18
e8,e9 8.2500 4
e1,e2,e3
3.7125 3
e1,e11,e9 1.6706 4
e1,e2,e12,e9 1.3669 4
T27
e2,e3,e4
13.7500
3
e11,e9,e4
6.1875
4
e11,e10,e5 2.7844 6
e2,e12,e9,e4 1.2530 4
e2,e12,e10,e5 1.0252 6
T47
e9,e4 11.5500 4
e10,e5 9.4500 6
T26 e11,e10 9.3500 6
e2,e12,e10 7.6500 6
T16
e7,e6 11.0000 8
e8,e10 4.9500 6
e1,e11,e10 4.0500 6
In Table1: the performance comparisons of bandwidth
allocation for paths are given. Every path set in a path is
analyzed separately using both algorithms.. The amount of
bandwidth allocated is tabulated respectively. The requested
bandwidth is distributed among the path sets according to the
respective algorithms.
In figure 3: and 4: the bandwidth allocation for path 1 and
path2 are given. For path1 the bandwidth allocation using
Max-Min fairness algorithm gives fair allocation. For path 2
the bandwidth allocation using Max-Min fairness algorithm
doesn’t meet the requirement. The GPAA exhibits much better
performance than the MMF in terms of bandwidth allocation
for path2.
In figure 5: the success rates of both the algorithms are
compared. The success rate is computed as the difference
between requested bandwidth and allocated bandwidth. The
success rate of GPAA is better compared to Max-Min fairness
algorithm. For path 2 and 3 the success rate is very less in
Max-Min fairness algorithm.
Figure 3: Illustrates bandwidth allocation for path1, where fairness is
obtained in MMF and priority analysis is seen in GPAA.
1 2 3 4 50
2
4
6
8
10
12
14
path set no
bandw
idth
allocation (
Mbps)
bandwidth allocation for path 2
MMF
GPAA
Figure 4: Illustrates bandwidth allocation for path 2; where in GPAA
the need is met and in MMA the allocation is not satisfied.
1 2 3 4 50.4
0.5
0.6
0.7
0.8
0.9
1
path no
success r
ate
(%)
path no VS success rate
MMF
GPAA
Figure5: the comparison of GPAA and MMF algorithm with success
rate. In figure 6: the price allocation for every path set of path 3 is
analyzed. For shortest path the price allocated is low.
ISBN-978-1-4507-2041-0 © by CiiT 2010 Published by Coimbatore Institute of Information Technology 4
Page 5
International Conference on Intelligent Information Systems and Management (IISM’2010), June10-12, 2010
1 2 3 4 50.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
path set no
price b
oundary
PRICE ALLOCATION FOR PATH 2
Figure 6: The satisfactory degree of path 2 through GPAA.
Taking 100 no of particles of size 10*10; price distribution
and utility distribution are computed as shown in figure 7: and
8: respectively.
0
5
10
0
5
100.5
1
1.5
resource no
Price distribution of 10 * 10 particles
user no
pric
e di
strib
utio
n
Figure 7: The particles distribution within the price boundary.
0
5
10
0
5
101
2
3
utility
dis
trib
ution
Utility distribution of 10 *10 particles
resource nouser no
Figure 8: The total utility value for 10*10 particles
In figure 8: the utility distribution of all the particles at the
final stage of GPAA execution is illustrated.
The routing optimality can be determined with the shortest
path and the feasible bandwidth allocated for that.
In Table 2: The performance comparisons of shortest path,
success rate, net bandwidth allotted are given. It illustrates that
the algorithm GPAA exhibit better performance than the
MMF algorithm in terms of bandwidth allocation and success
rate, whereas they have approximately same shortest paths.
TABLE 2: OPTIMAL ANALYSIS OF PARAMETERS
Path
Requested
Bandwidth
Network
Bandwidth
Utilization
Success rate Shortest path
GPAA MMF GPAA MMF GPAA MMF
T18 15 15 15 100% 100% e8,
e9
e8,
e9
T27 25 25 23 100% 92%
e2,
e3,
e4
e11,
e9,
e4
T47
21
21
10
100%
47%%
e9,
e4
e10,
e5
T2 6
17
17
12
100%
70%
e11,
e10
e11,
e10
T16
20
20
20
100%
100%
e7,
e6
e7,
e6
VI. CONCLUSION
In this paper a generalized particle approach algorithm is
used to estimate the bandwidth allocation in ATM networks.
The performance evaluation using GPAA is compared with
MMF algorithm and found better in terms of success rate,
network bandwidth utilization, price allocation and quality of
service. In future, congestion factor, breakdown factor,
fairness factor can be incorporated in GPA model to optimize
the bandwidth allocation.
REFERENCES
[1] Dianxun Shuai, Xiang Feng, Francis C.M.Lau, A new generalized
Particle approach to parallel bandwidth allocation. Computer
Communications July 2006 3933-3945.
[2] Dianxun Shuai, Yuming Dong, Qing Shuai, Optimal control of Network
services based on Generalised particle model.
[3] Saswati Sarkar, Leandros Tassiulas,Fair Distributed congestion control
in multirate multicast networks IEEE/ACM Transactions on Networking
vol.13.no1, February 2005
[4] Dimitri Bertsekas, Robert Gallager, Data Networks. Englewood cliffs
NJ: Prentice Hall,1987
[5] Ammar W.Mohammed and Nirod Chandra sahoo, Efficient
computation of shortest paths in networks using Particle Swarm
ISBN-978-1-4507-2041-0 © by CiiT 2010 Published by Coimbatore Institute of Information Technology 5
Page 6
International Conference on Intelligent Information Systems and Management (IISM’2010), June10-12, 2010
optimization and Noising metaheuristics ,Hindawi publishing
Corporation ,volume 2007,Article Id 27383,
[6] Bozidar Radunovic,A unified framework for Max-Min and Min-max
Fairness with applications,july2002.
[7] Chang Wook Ahn,R.S.Ramakrishna, A Genetic Algorithm for Shortest
Path routing problem and the sizing of populations. IEEE/
Transactions on Evolutionary Computation, vol, 6, N0:6, December
2002.
[8] Particle Swarm optimization Tutorial
www.swarmintelligence.org/tutorials
ISBN-978-1-4507-2041-0 © by CiiT 2010 Published by Coimbatore Institute of Information Technology 6