Вы находитесь на странице: 1из 8

Optical Fiber Technology 18 (2012) 490–497

Contents lists available at SciVerse ScienceDirect

Optical Fiber Technology


Regular Articles

Traffic grooming in WDM optical network with grooming resources at Max

Connectivity nodes
Partha Paul a, Balbeer Singh Rawat a, S.K. Ghorai b,⇑
Department of Computer Science, Birla Institute of Technology, Mesra, Ranchi 835 215, India
Department of Electronics and Communication Engg., Birla Institute of Technology, Mesra, Ranchi 835 215, India

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, we propose Max Connectivity grooming in WDM mesh networks under static lightpath con-
Received 2 April 2012 nection requests. The grooming and wavelength conversion resources are placed at the nodes having
Revised 23 June 2012 maximum connections. We propose a heuristic genetic algorithm (GA) model to solve grooming, routing
Available online 3 September 2012
and wavelength assignment. The GA algorithm has been used to optimize the cost of grooming and wave-
length conversion resources. The blocking probability has been investigated under different lightpath
Keywords: connections. The performance of Max Connectivity grooming has been compared with other grooming
WDM optical networks
policies. Our results indicate the improvement of resource utilization with minimum blocking
Wavelength assignment
Traffic grooming Ó 2012 Elsevier Inc. All rights reserved.

1. Introduction nel of high bandwidth. The grooming node includes optical cross
connects (OXCs), demultiplexers (DMUXs), multiplexers (MUXs)
Optical network with wavelength division multiplexing (WDM) and/or digital cross connects (DXCs) that perform grooming as well
has been an active area of research, as it can meet the huge band- as switching operations. These grooming resources efficiently mul-
width requirement of internet and telecommunication data traffic tiplex several low speed traffic streams on to high capacity wave-
through WAN, MAN or LAN. In WDM, multiple wavelengths are length channels and also demultiplex those whenever required. In
transmitted through a single fiber and the transmission capacity addition to grooming, these devices add connections, drop connec-
can be achieved as high as hundreds of terabit-per-second. Each tions and also perform wavelength conversion using add drop mul-
wavelength can support a data rate of several gigabit per second tiplexers (ADMs) and wavelength converters (WCs).
(e.g. OC-192, OC-768, etc.). However, in practical networks, the In the beginning, the traffic grooming problems were imple-
traffic requested by an individual connection is in the range of mented in ring networks [2–7], and later on, it has been focused
megabits-per-second. Hence, a significant portion of transmission on WDM mesh networks [1,8–12]. In WDM mesh networks, the
capacity of a wavelength channel would be wasted. For effective traffic scenarios can be static, incremental or dynamic [13]. Also
utilization of the bandwidth, traffic grooming mechanism is being the traffic grooming can be single hop traffic grooming or multi
used in the network where low speed traffic streams are multi- hop traffic grooming depending upon the lightpaths used by a con-
plexed or groomed into high speed wavelength channels [1]. It nection between source and destination. The traffic grooming
minimizes the network cost by reducing the number of ADMs problem has been addressed in WDM mesh networks by several
and the number of wavelengths used in the network. In WDM net- groups of researchers for different traffic conditions. However,
work, a lightpath must be established to carry traffic using a partic- our work focuses on static traffic grooming in WDM mesh network.
ular wavelength and proper route between a source node and a To maximize the network throughput, an integer linear program-
destination node. Although this routing and wavelength assign- ming (ILP) formulation was proposed in a six nodes irregular mesh
ment problem (RWA) minimizes the connection blocking, but it network [1]. All the nodes have grooming capability and heuristic
cannot utilize the network resources effectively. approaches were used to solve traffic grooming problem in large
Traffic grooming with RWA resolves the network bandwidth networks. A multi-objective evolutionary algorithm was reported
problem more effectively. It provides provisioning connection re- to optimize multiple objectives at the same time in WDM mesh
quests of different bandwidth granularities on to wavelength chan- network [14]. It maximizes the throughput and minimizes the net-
work cost and average hop counts. A decomposition method based
⇑ Corresponding author. Fax: +91 651 2275401. an integer linear programming (ILP) was used in a large WDM
E-mail addresses: p_india@rediffmail.com (P. Paul), skghorai12@rediffmail.com
mesh network [15]. It divides the traffic grooming, routing and
(S.K. Ghorai). wavelength assignment (GRWA) problem into two parts. One part

1068-5200/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved.
P. Paul et al. / Optical Fiber Technology 18 (2012) 490–497 491

Fig. 1. (a) 16-Node WDM mesh network where (1, 2, 7, 8, 9 and 16) are edge nodes and (5, 12 and15) are maximum connectivity nodes and (b) 20-node WDM mesh network
where (1, 2, 3, 10, 16, 17, 18 and 19) are edge nodes and (2, 7, 14, 18 and19) are maximum connectivity nodes.

is traffic grooming and routing problem and the other one is wave- grated auxiliary graph has been proposed to reduce the cost and
length assignment problem. It produces an optimal solution under enhance resource utilization in WDM networks [21]. Multi-hop
two considerations, one wavelength capacity constraint and the grooming and waveband switching are integrated to decrease the
other one wavelength continuity constraint. Another decomposi- blocking probability and to save the network cost.
tion method was reported where a large network is considered In the present work, we propose the Max Connectivity groom-
as a group of clusters [16]. A greedy heuristic algorithm has been ing in mesh network to solve GRWA problem under static lightpath
used to groom intra-cluster traffic within each cluster. In order to request. In most of the previous works, grooming devices were
reduce the network cost without affecting the network perfor- placed all over the optical network except [17] where grooming re-
mance, edge grooming was proposed [17], where grooming re- sources are distributed at the edges of the network. Our study re-
sources are placed on the edge of optical networks instead of veals that placing the grooming devices at the nodes having
distributing throughout the network. The most contiguous heuris- maximum connection is more cost effective than all grooming
tic algorithm was used to solve grooming problem without wave- and edge grooming, and results similar blocking performance as
length conversion resources and the genetic algorithm was used achieved with edge grooming. Earlier, the maximum connectivity
with wavelength conversion resources. An algorithm based on var- grooming has been used in unidirectional path switch ring network
iant load balancing under the hose uncertainty model was investi- [22]. In our work, genetic algorithm (GA) has been used to solve
gated for multi-granularity WDM mesh networks [18]. It has been grooming and wavelength assignment (GRWA) problem in WDM
shown that minimizing path cost first performs better than maxi- mesh network. The GA procedure is simple and easily applicable
mize resources utilization first. to large networks. It works on a large solution set (search space)
A fixed order multi-hop traffic grooming based on fixed alter- which makes it more efficient than other approaches such as ILP,
nate routing has been used to address grooming node selection greedy heuristics approach, most contiguous heuristic algorithm
in WDM optical network without wavelength conversion capabili- and other heuristics proposed in the literature. In ILP, computation
ties [19]. Unlike the previous decomposition approaches, a multi- complexity increases with the increase in the size of the network
level decomposition approach which decomposes traffic at four due to the increase in the number of variables. Also all traffic re-
different levels has been proposed to evaluate the blocking perfor- quests must be known in advance for solving GRWA problem. Heu-
mance. Recently, a heuristic algorithm called waveband grooming ristic approaches are faster as compared to GA. However those are
with layered auxiliary graph has been reported in multi domain based on observations which may be invalid over time. Also, they
optical networks to improve the scalability and to reduce the net- are applied to some specific type of networks. Although greedy
work cost [20]. The network is divided into two layers. The high algorithm can be used for variety of networks, but it does not pro-
layer is the virtual topology layer which includes inter-domain vide optimal solution with added constraints.
routing between the source node and destination node and the In Section 2, the formulation of GRWA problem with static traf-
low layer includes the intra-domain routing in each single domain. fic requests has been presented. Section 3 describes the GA ap-
A new multi-granularity traffic grooming algorithm based on inte- proach to solve GRWA problem with maximum connectivity
492 P. Paul et al. / Optical Fiber Technology 18 (2012) 490–497

grooming. In Section 4, we present performance with different The general cost function given by Eq. (1a) also considers that
lightpath requests and cost comparison for three different groom- wavelength conversion devices are used with the grooming
ing techniques followed by conclusion in Section 5. devices.
Eq. (1a) is subject to the following constraints:
2. Network architecture and problem formulation
i. At most one lightpath can be setup between two nodes using
To demonstrate GRWA problem with maximum connection a single wavelength,
grooming, we have considered two different network architec-
tures: (i) a 16-node WDM mesh network as shown in Fig. 1a, the pw
i;j 6 bði; jÞ ð1bÞ
nodes 5, 12, and 15 are considered with maximum connection i;j
and the nodes 1, 2, 7, 8, 9, and 16 are considered as edge nodes
where w is the particular wavelength used between node i
for comparison, and (ii) a 20-node mesh network as shown in
and node j, pw
i;j implies that wavelength w has been assigned
Fig. 1b, where 2, 7, 14, 18 and 19 are maximum connectivity nodes,
to the lightpath from node i to node j, variable b(i, j) shows
and 1, 2, 3, 10, 16, 17, 18 and 19 are edge nodes.
whether lightpath from node i to node j exists or not.
We have made the following assumptions for solving the GRWA
ii. Grooming and wavelength conversion capability of each
node is limited to the number of grooming devices placed
on that node.
i. All links are bidirectional links.
iii. Grooming state of each grooming device is limited to the
ii. Grooming and wavelength conversion resources should be
maximum capacity of the fiber link, i.e.
placed on the nodes having maximum connectivity.
iii. Capacity of one wavelength on one fiber is constant ‘C’ (OC- X
48). yci;j 6 C max ð1cÞ
iv. Traffic demand is static. Connection requests (R) are known
in advance. where r e L (lightpath requests), c = [1, 3, 12, . . ., Cmax], Cmax is the
v. Number of wavelengths available per fiber is limited (we maximum capacity of the fiber link, yci;j is the lightpath request be-
have used eight wavelengths per fiber). tween node i and j with bandwidth c.
vi. Traffic requests may be through any one of speeds: OC-1, In the present work, we have considered that the nodes will be
OC-3, OC-12 and OC-48. equipped with both grooming as well as wavelength conversion re-
sources. Without using wavelength conversion, wavelength conti-
2.1. Network cost formulation nuity constraint is to be fulfilled. As a result more wavelengths are
to be used for establishing the connection requests which is not
The major cost of the optical networks is due to the wave- cost effective. In our work, we have minimized the objective func-
lengths used in connections establishment, and grooming and tion given by Eq. (1a). We can minimize the objective function
wavelength conversion devices deployed on the nodes. The con- either by minimizing the number of devices used (grooming and
nection requests are static, i.e. all the requests are known in ad- wavelength conversion devices) or by minimizing the number of
vance. Let N be the number of nodes in the mesh network in hop count Hi. We have used max-connectivity grooming and have
which grooming and wavelength conversion resources are placed, succeeded in minimizing the blocking probability and simulta-
L be the requested connections, CG be the grooming cost and CW neously reducing the cost of the network by minimizing the usage
be the wavelength conversion cost. So the objective function for of grooming and wavelength conversion devices.
cost can be considered as:
2.2. Max Connectivity grooming
Total cost ¼ fðai  CGÞ þ ðbi  CWÞg þ Hi ð1aÞ
In this section, we have shown how max-connectivity grooming
i¼0 i¼0
can be effective in reducing the cost and improve the blocking per-
where ai is the number of grooming devices used at the ith node, bi formance. Fig. 2a and b displays two optical mesh networks having
the number of wavelength conversion resources used at the ith two wavelengths available in each link and all the links are bidirec-
node, Hi the number of hops used by the ith connection request. tional. The maximum capacity of each wavelength is assumed to be

Fig. 2. (a) Grooming devices placed at node 3 and node 5. (b) Grooming devices placed at node 1 and node 4.
P. Paul et al. / Optical Fiber Technology 18 (2012) 490–497 493

OC-12. We have placed grooming devices on the nodes having c. Mutation operator.
maximum connectivity as shown in Fig. 2a. In Fig. 2b we have
placed grooming devices randomly on nodes 1 and 4. INITIAL POPULATION (chromosomes formation): Initial popula-
Suppose we have six connection requests of OC-3 as described tion is formed with the help of all the possible paths for each light-
in Table 1. path requested. For the 16-node mesh network of Fig. 1a the initial
Using the arrangement of grooming devices in Fig. 2a and b, we population for lightpath request 1–14 can be as follows (Table 2):
have assigned the resources to the connection requests as depicted
in Table1. Two wavelengths (k1, k2) are available per fiber. It may be
3.1.1. Selection operator
seen that using the device arrangement of Fig. 2a all the connection
Selection of an individual route is done with the help of a fitness
requests are successfully established. On the other hand using
function. In our research work, the fitness function is inversely pro-
Fig. 2b device arrangements, connection request 2 ? 6 cannot be
portional to the cost function that we described earlier in Section
established. Either we can use one more wavelength or we can de-
2.1 by Eq. (1a).
ploy a grooming device on node 6. This will increase the cost. So, it
The fitness function estimates the fitness of each individual
may be seen that the proper placement of grooming devices on the
with respect to the cost calculated using the cost function given
nodes reduces the cost of the optical networks.
by Eq. (1a). The GA procedure calculates the cost associated with
the routes as the sum of the individual costs due to the wave-
3. GA approach to solve the GRWA problem lengths and devices used.
Each individual route is selected on the basis of the fitness it
Genetic algorithms (GAs) [23] are search algorithms based on holds. Routes having fitness less than the threshold value would
the mechanics of natural selection and natural genetics. A GA be discarded.
works with chromosomes, each of which represents a solution to The cost function described earlier by Eq. (1a) is used for the
the problem. A fitness function is defined in order to estimate optimization. All the chromosomes in the population will have a
the probability of a solution for being selected. An initial popula- fitness value assigned to it and this fitness value is calculated using
tion of individuals is created and then evolved by means of genetic the cost function. The fitness value assigned to each chromosome is
operators, such as crossover and mutation, to form a new popula- based upon the number of devices and the number of wavelengths
tion (the next generation) that is hoped to be fitter than the last used in the connection establishment.
one. The goodness of the new set of solutions is determined with The higher the cost of route, the lower will be the fitness value,
the help of the fitness function. The evolution process is repeated so the route with minimum cost will be selected for connection
a predefined number of iterations or until some other criterion is establishment. After each iteration of our proposed GA procedure,
met. all chromosomes (established paths) will have a fitness value as-
Several researchers have proposed GA solution to GRWA prob- signed to them. Let the chromosomes are represented as Chr = Pab
lem earlier [17,24,25]. Most of them have considered a full groom- , . . ., Pij where Pij represents the path between the source i and des-
ing network and have decomposed the GRWA problem in three sub tination j. Suppose GRD(Pij) and WCD(Pij) represent the number of
problems: grooming, routing and wavelength assignment. We grooming and wavelength conversion devices used in the path Pij
have solved GRWA problem as a single problem. respectively. Then the cost of each individual chromosome is given
3.1. Description of genetic operators used

The following genetic operators have been used in our GA CðChrÞ ¼ ðWLi;iþ1 Þ þ C½GRDðPij Þ þ WCDðPij Þ ð2Þ
where n is the number of nodes in the path Pij and WLi,i+1 is the
a. Selection operator. wavelength used between nodes i and i + 1. C[GRD(Pij) + WCD(Pij)]
b. Crossover operator. is the cost of resources used in the path Pij.
Now all the chromosomes are sorted in ascending order with
respect to their cost, so that the chromosome with the least cost
Table 1 is chosen. We have also calculated the cumulative fitness of the
Connection requests and wavelength assignment to each request.
chromosomes after each iteration. This cumulative fitness is given
s–d No. of Wavelength Wavelength by:
connections assignment for Fig. 2a assignment for Fig. 2b
requests X

1?4 1 k1 k1
CMFðChrÞ ¼ CðChri Þ ð3Þ
5?4 1 k2 k2 i¼0

5?3 2 k1, k1 k1, k2

2?6 1 k2 – where LPR is the number of lightpath requests and C(Chri) is the
2?4 1 k1 k1 cost of ith chromosome.
The cumulative fitness (CMF) is used to restrict the number of
iteration in the GA algorithm. For maximum connection establish-
ment the individual fitness of the chromosomes should be best. So
Table 2 after each iteration of the GA algorithm current fitness is compared
Set of possible paths for connection request 1-14.
with the previous fitness. If the fitnesses are equal then the execu-
1 7 11 12 13 15 14 tion is halted and the result obtained is the optimal result.
1 7 11 12 10 15 14
1 6 5 12 13 15 14
1 2 3 4 8 9 14 3.1.2. Crossover operator
1 2 3 4 8 10 15 14 Crossover operator can only be applied to the routes (chromo-
1 6 5 4 8 10 12 13 15 14
somes) having one or more common nodes except the source
1 7 11 12 5 4 8 9 16 15 14
and the destination node. Taking the mesh network in Fig. 1a the
494 P. Paul et al. / Optical Fiber Technology 18 (2012) 490–497

Fig. 3. Crossover between two routes.

Fig. 4. Formation of new route after mutation.

two routes from node 1 to node 14, P0 and P1 may be used as two
parent chromosomes to perform crossover.
In Fig. 3, parent routes P0 and P1 are used to form two new
routes (offspring) C0 and C1. The crossover point is the common
node in the parent routes, i.e. node 5. The new routes are selected
to form new population on the basis of the fitness value. The higher
is the fitness value, the greater is the chance of getting selected. If
crossover is not performed, the offsprings are the exact copy of the
parent routes.

3.1.3. Mutation operator

A mutation is a permanent change in the sequence of the route.
Mutation is applied to the routes (chromosomes) having fitness va-
lue lower than the threshold value. Threshold value is chosen as
the mean fitness value of all the routes in the population. A random
node is chosen as the mutation point and it must have connectivity
more than two, so that a different path can be discovered.
In the underlying example (Fig. 4), we have used the mesh net-
work as shown in Fig. 1a and taken a route from node 1 to node 9.
The mutation point is selected as node 12. After mutation, the new Fig. 5. Blocking probability vs. traffic grooming and wavelength conversion
route’s fitness is evaluated and if its fitness value is greater than resources.
the threshold then the new route is added to the new population.

3.2. GA procedure

Outline of the procedure used: The blocking probability for a specified number of lightpath re-
quests is determined as follows:
 Create AdjMat(); ðTotal lightpath requests  Successfully established pathsÞ
 Create Load(); Blocking probability ¼  100
Total lightpath requests
 Device Order();
 Creating and initializing Chromosomes:
 For GenLimit > 0 4. Results and discussion
 For LPR > 0
 Path selection and wavelength assignment The performance of our proposed GA based heuristic procedure
 Checking fitness. to solve GRWA problem has been demonstrated in WDM optical
 Cross over() 16-node and 20-node network topologies as illustrated in Fig. 1a
 Mutation() and b. We have compared our results with edge grooming and
 Checking fitness again and adding offspring to the all grooming. We have found that it is efficient and cost effective
population. to deploy grooming and wavelength resources on the nodes having
 Use new generated population for a further run of maximum connectivity instead of placing resources randomly over
algorithm. the network.
 Return number of successful paths. In our work, we have considered that the wavelength conver-
 Calculation of blocking probability. sion devices are only placed on the nodes with grooming capability
(as traffic grooming devices are capable of performing wavelength
P. Paul et al. / Optical Fiber Technology 18 (2012) 490–497 495

Fig. 6. Blocking probability vs. number of lightpath requests: (a) for 16-node mesh network and (b) for 20-node mesh network.

Fig. 7. Blocking probability vs. traffic grooming and wavelength conversion resources using 150 lightpath requests: (a) for 16-node mesh network and (b) for 20-node mesh

Fig. 8. Total cost vs. number of connections: (a) for 16-node mesh network and (b) for 20-node mesh network.
496 P. Paul et al. / Optical Fiber Technology 18 (2012) 490–497

conversion too). There are eight wavelengths available per fiber in imum connectivity and for 20-node network, we have used nodes
the network and the algorithm has been executed until the cumu- 2, 7, 14, 18, and 19 having maximum connectivity. For max-con-
lative fitness is found equal for consecutive generations (optimal nectivity grooming in 16-node network, we place resources only
result). Single point crossover is done and the mutation probability on nodes 5, 12, and 15 and for edge grooming, we have assumed
is 1/n, where n stands the number of nodes in the chromosome. that only nodes 1, 2, 7, 8, 9, and 15 would be equipped with the
In Fig. 5, blocking probability for 75, 100 and 125 lightpath re- grooming and wavelength conversion resources. For max-connec-
quests is plotted with traffic grooming and wavelength conversion tivity grooming in 20-node network, we place resources only on
resources. The results are shown for the 16-node network illus- nodes 2, 7, 14, 18, and 19 and for edge grooming, we have assumed
trated in Fig. 1a. Initially the blocking probability for each of the re- that only nodes 1, 2, 3, 10, 16, 17, 18, and 19 would be equipped
quests is very high and it reduces as we increase the resources in with the grooming and wavelength conversion resources. The eval-
the network. uation is done for different connection requests, e.g. 50, 75, 100,
Fig. 5 shows that as the lightpath requests increase, the blocking 125, 150, 200, 300 lightpath requests and for each policy only
probability of the network also increases. It can be observed from two grooming and wavelength resources are used. The result
Fig. 5 that for two grooming and wavelength conversion resources, shows that placing devices on the nodes having maximum connec-
blocking probability for 125 lightpath requests is much higher than tivity improves the blocking probability than the other policies.
100 and 75 lightpath requests. Placing the devices randomly on the nodes results in a high block-
In Fig. 6, results of three grooming policies are compared and ing probability where as using max-connectivity grooming policy
the results are shown for two different network architectures. In results in lower blocking probability than the others.
this evaluation, the resources placed on the nodes are fixed and Our results show that max-connectivity grooming policy is bet-
then for each grooming policy the results are compared. For 16- ter than the other policies in terms of blocking probability and also
node mesh network, we have used nodes 5, 12, and 15 having max- it is cost effective. Also, it may be observed that the blocking prob-

Fig. 9. Blocking probability vs. blocked connections: (a) connection failure for all grooming, (b) connection failure for max-connectivity grooming, and (c) connection failure
for edge grooming.
P. Paul et al. / Optical Fiber Technology 18 (2012) 490–497 497

ability for 20-node mesh network is less as compared to 16-node wavelength conversion resources are deployed in the nodes having
mesh network. This is due to the presence of more number of links maximum connectivity. A heuristic genetic algorithm model has
in the 20-node network as compared to 16-node network. been used to solve the traffic grooming under static lightpath re-
Fig. 7 shows a comparison between the results obtained for two quests. It has been seen that the Max-Connectivity grooming im-
grooming policies from our genetic procedure using 16-node and proves the blocking probability compared to all-grooming
20-node mesh network architectures. The results shown in Fig. 7 (devices placed randomly) and edge-grooming for a fixed number
are conducted for 150 lightpath requests. of lightpath requests. Also it is more cost effective for a certain
In our GA based heuristic approach to solve the GRWA problem, number of connections. Our work can be extended to WDM mesh
the grooming devices are increased additively and the resultant networks under dynamic lightpath requests with the distributed
blocking probability is examined. As the resources are increased grooming and wavelength conversion resources at the nodes hav-
in the network the number of successfully established paths also ing maximum connectivity.
increases and hence improves the blocking probability. Our results
show that the performance of the max-connectivity grooming is References
comparable to the edge grooming policy for both the network
architectures. [1] K. Zhu, B. Mukherjee, Traffic grooming in an optical WDM mesh network, IEEE
J. Select. Areas Commun. 20 (1) (2002) 122–133.
Fig. 8 shows the comparison of cost for three different grooming [2] R. Dutta, G.N. Rouskas, On optimal traffic grooming in WDM rings, IEEE J.
policies using two different network architectures (16-node and Select. Areas Commun. 20 (1) (2002) 110–121.
20-node mesh networks). The network cost is measured by the [3] A.L. Chiu, E.H. Modiano, Traffic grooming in algorithms for reducing electronic
multiplexing costs in WDM ring networks, IEEE/OSA J. Lightw. Technol. 18 (1)
wavelength link cost, grooming and wavelength conversion de- (2000) 2–12.
vices cost used in the network. The cost shown in Fig. 8 is the nor- [4] J. Wang, V.R. Vemuri, W. Cho, B. Mukherjee, Improved approaches for cost
malized cost depending upon the objective function given by Eq. effective traffic grooming in WDM ring networks: ILP formulations and single-
hop and multi-hop connections, IEEE/OSA J. Lightw. Technol. 19 (11) (2001)
(1a). The cost is evaluated as: 1653–1654.
[5] X. Zhang, C. Qiao, An effective and comprehensive approach for traffic
ðNumber of hopsÞ units of link cost grooming and wavelength assignment in SONET/WDM rings, IEEE/ACM
þ ðNumber of grooming devicesÞ  Cost of single device Trans. Network. 8 (5) (2000) 608–617.
[6] O. Gerstel, R. Ramaswami, G.H. Sasaki, Cost-effective traffic grooming in WDM
þ ðNumber of wavelength conversion devicesÞ rings, IEEE/ACM Trans. Network. 8 (5) (2000) 618–630.
 Cost of single device: [7] E. Modiano, P.J. Lin, Traffic grooming in WDM networks, IEEE Commun. Mag.
39 (2001) 124–129.
We have evaluated the performance for a constant blocking [8] S. Thiagarajan, A.K. Somani, Capacity fairness of WDM networks with
grooming capabilities, Opt. Networks Mag. 2 (3) (2001) 24–31.
probability (i.e. 15%) for all the grooming policies. The result shows [9] L.A. Cox, J. Sanchez, Cost savings from optimized packing and grooming of
that placing the devices on the nodes with maximum connectivity optical circuits: mesh versus ring comparisons, Opt. Networks Mag. 2 (3)
improves the cost factor. Here we can see that the cost increases (2001) 72–90.
[10] C. Xin, C. Qiao, S. Dixit, Traffic grooming in the mesh WDM optical network –
with the increase in the number lightpath requests. As the connec- performance analysis, IEEE J. Select. Areas Commun. 22 (9) (2004) 1658–1669.
tion requests increases, the resources available over the network [11] H. Zhu, H. Zang, K. Zhu, B. Mukherjee, A novel generic graph model for traffic
must also be increased to satisfy as much as lightpath requests grooming in heterogeneous WDM mesh networks, IEEE/ACM Trans. Network.
11 (2) (2003) 285–299.
possible. As a result the number of wavelengths used is also in- [12] C. Xin, B. Wang, X. Cao, J. Li, Logical topology design for dynamic traffic
creased and more importantly it increases the cost factor. In grooming in WDM mesh networks, J. Lightw. Technol. 24 (2006) 2267–2275.
Fig. 8a and b, it is depicted that to achieve the blocking probability [13] A.E. Ozdaglar, D.P. Bertsekas, Routing and wavelength assignment in optical
networks, IEEE/ACM Trans. Network. 11 (2) (2003) 259–272.
of 15% for different lightpath requests (50, 75, 100, etc.) max-con- [14] P. Prathombutr, J. Stach, E.K. Park, An algorithm for traffic grooming in WDM
nectivity grooming policy is most cost effective than the other pol- optical mesh networks with multiple objectives, in: Proceedings of IEEE
icies. The results also indicate that, increasing the nodes in the International Conference on Computer Communications and Networks, 2003,
pp. 405–411.
network also increases the total network cost.
[15] J.Q. Hu, B. Leida, Traffic grooming, routing and wavelength assignment in
Fig. 9 shows the failure rate comparison between the different optical WDM mesh networks, in: Proceedings of IEEE INFOCOM, 2004, pp.
grooming policies and the bars in the graph represent the blocked 495–501.
connections. In all grooming, all the 16 nodes will be equipped [16] S. Huang, R. Dutta, G.N. Rouskas, Traffic grooming in path, star, and tree
networks: complexity, bounds, and algorithms, IEEE J. Select. Areas Commun.
with grooming capability where as in max-connectivity grooming 24 (4) (2006) 66–82.
only nodes 5, 12, and 15 will have grooming capability and edge [17] O. Awwad, A.I. Al-Fuqaha, A. Rayes, Traffic grooming, routing and wavelength
grooming only nodes 1, 2, 7, 8, 9, and 16 will have grooming capa- assignment in WDM transport networks with sparse grooming resources,
Comput. Commun. 30 (2007) 3508–3524.
bility. All the plots show nth connection failure for different traffic [18] X. Zhang, L. Li, S. Wang, X. Wei, A novel robust routing algorithm for multi-
grooming techniques for 150 lightpath requests. It can be observed granularity connection requests in wavelength division multiplexing mesh
that most of the initial connection requests are established due to networks, Opt. Fiber Technol. 14 (2008) 119–129.
[19] W. Yao, G. Sahin, M. Li, B. Ramamurthy, Analysis of multi-hop traffic grooming
the availability of the resources. As the resources are limited, more in WDM mesh networks, Opt. Switch. Network. 6 (2009) 64–75.
blocking is observed when lightpath request increases. [20] L. Guo, J. Cao, X. Wang, D. Jiang, T. Yang, Waveband grooming based on layered
In all grooming, the number of failed connections is less as com- auxiliary Graph in multi-domain optical networks, Opt. Fiber Technol. 16
(2010) 162–171.
pared to max-connectivity and edge grooming, but the cost is very [21] X. Wang, W. Hou, L. Guo, J. Cao, D. Jiang, A new multi-granularity grooming
high due to the placement of grooming devices in all the nodes. The algorithm based on traffic partition in IP over WDM networks, Comput.
max-connectivity and edge grooming shows quite similar perfor- Networks 55 (2011) 807–821.
[22] Y. Wang, Qian-Ping Gu, Maximizing throughput for traffic grooming with
mance as compared to all grooming using less number of
limited grooming resources, IEEE GLOBECOM (2007) 2337–2341.
resources. [23] David E. Goldberg, Genetic Algorithms in Search, Optimization and Machine
Learning, Pearson Education, 2006.
[24] C. Lee, E.K. Park, A genetic algorithm for traffic grooming in all-optical mesh
5. Conclusions networks, IEEE Int. Conf. Syst. Man Cybernet. 7 (2002) 6–9.
[25] R.S. Barpanda, A.K. Turuk, B. Sahoo, A new cost function to solve RWA problem
We have demonstrated traffic grooming, routing and wave- in wavelength routed optical network using genetic algorithms, World Congr.
Nat. Biologic. Inspir. Comput. NaBIC (2009) 1297–1302.
length assignment in WDM mesh network where grooming and