060.

4256 Networks, network optimization

Connection Admission Grooming in PCE Based Architecture incorporating Dynamic Fiber State Information with combined RWA
Sidharth Shukla, Member IETE, Vimal Bhatia, Senior Member IEEE
Abstract- Over the years telecommunication networks have undergone huge technological advancements emerging from circuit switched networks to state of the art optical networks. At present, pure optical wavelength routed networks have taken prominence as the future backbone telecommunication networks, incorporating technologies like dense wavelength division multiplexing and generalized multiprotocol labeled switching for its control plane. To efficiently optimize these networks, efficient traffic engineering, dynamic routing and wavelength assignment techniques need to be incorporated to improve connection blocking probabilities. The proposed connection admission grooming algorithm implements the above in an efficient manner. It is shown that the proposed algorithm reduces the blocking probability of the network, thus resulting in substantial improvement in network resource utilization. employed throughout the world for backbone communication requirements. Such high data rate, of the order of tens of thousands of gigabits per second, carried by a single strand of fiber seems sufficient for the present requirements. However with the telecommunication, cellular and internet user base increasing in an exorbitant manner around the world [2], soon this bandwidth will seem meager in comparison to backbone data rate requirements of the future. The entire paper has been organized into six sections. The second section defines the problem. Wavelength routing in optical networks is discussed in the third section. The fourth section explains the Generalized Multi Protocol Label Switching (GMPLS) based optical networks. The proposed connection admission grooming algorithm is explained in the fifth section. The sixth and final section analyses the results along with their discussions.

Index Terms- Multi Protocol Label Switching (MPLS),

Wavelength Division Multiplexing (WDM), Generalized Multi Protocol Label (GMPLS), Routing and Wavelength Assignment (RWA), Resource Reservation Protocol- Traffic Engineering (RSVP-TE), and Path Computation Element (PCE).

II. Problem Statement

A requirement is perceived to optimally utilize the bandwidth provided by optical networks. This can be done by efficient routing and wavelength assignment techniques coupled with optimal traffic engineering. Though various studies and priority based algorithms for traffic grooming in optical networks have recently been undertaken [3] - [6], but the aspect of prioritization of user traffic to support Quality of Service (QoS), enforcing Service Level Agreements (SLAs) and use of fiber link state information has not been incorporated in these algorithms. The proposed Connection Admission Grooming (CAG) algorithm addresses all the above issues and has effectively incorporated dynamic fiber state information, class of service quantization and prioritization of service requests. Path computation element based architecture has been incorporated to facilitate routing and signaling requirements of complex optical networks. The proposed CAG algorithm can effectively groom and prioritize the service traffic to improve the overall connection blocking probability of

I. Introduction
Over the years telecommunication networks have evolved considerably, from capacity or bandwidth enhancements to improvements in speeds and reduction in costs. Optical networks employing pure optical regeneration provide high data rates/speeds in the backbone network [1]. Wavelength Division Multiplexing (WDM) technique employed with optical amplifiers could multiplex high number of wavelengths (or communication channels) on to a single strand of fiber. Thereafter variations to WDM in the form of Coarse Wavelength Division Multiplexing (CWDM) and Dense Wavelength Division Multiplexing (DWDM) techniques were conceived. Due to the above reasons DWDM networks have become the backbone network of the future and

060.4256 Networks, network optimization

the network in highly utilized optical networks, where the effectiveness of the algorithm is substantial. The simulation results have demonstrated that incorporating the proposed CAG algorithm with combined routing and wavelength assignment in path computation element based architecture, the connection blocking probability, which is an important parameter for judging the efficiency of any network, reduces substantially. This in turn substantially increases the utilization of network resources and consequently improves the efficiency of pure optical networks. The CAG algorithm has better connection blocking performance than the Priority based algorithm [6] which uses only two variables viz. number of hops and data rate of requested service as compared to CAG algorithm which incorporates four variables viz. class of service, number of hops, data rate and fiber state to calculate priority for each service request. The shortcomings of the Priority based algorithm [6] have also been analyzed.

2
assignment problem. There are various methodologies for wavelength routing ranging from fixed routing, fixed-alternate routing and adaptive routing. The wavelength assignment techniques can be static or dynamic. In static wavelength assignment, the wavelength assignment is done offline for all the lightpaths which could be required for the operation of the network. The objective of static wavelength assignment is to accommodate all demands of lightpaths, while reducing the sum of all the wavelengths used in all links of the optical network. Static wavelength assignment becomes complex for large networks and hence for commercial networks dynamic wavelength assignment techniques are used. For dynamic wavelength assignment situation, the lightpaths required are not known initially as in static wavelength assignment, but here lightpath requests arrive independent of each other and in a random fashion. The objective here is to minimize the connection blocking probability over the entire network. Various heuristic [8] are available with dynamic wavelength assignment, they are - Random Fit, First Fit, Least Used, Most Used, Min Product, Least Loaded, Max Sum and Relative Capacity Loss.

III. Wavelength Routed Optical Networks

All optical networks define lightpath [7] as a channel used for communication. The lightpath is established between an optical node having information and an optical node desirous of that information, and may be interlinked by multiple such nodes. These nodes may be organized in a meshed architecture and the lightpath uses a single wavelength in every link it traverses. These optical nodes route the wavelength of the lightpath, using optical switches, by the best and shortest route possible as calculated by the algorithm in the network. The wavelength in every link of the lightpath may be same or different. We will assume that wavelength convertors are not used in the network, as it degrades the quality of the signal at each convertor employed. But this brings us to another complex problem which is to ensure that every lightpath in the optical network should have the same wavelength in all intermediate links from originating node to destination node. This requirement is termed as Wavelength Continuity Constraint (WCC), i.e., the wavelength will be continuous (it will not be converted by any of the intermediate nodes) from initial to the last link of every lightpath. This constraint further has a requirement to ensure that on every optical link, having multiple lightpaths, should have distinct wavelengths assigned to them to ensure WCC. Hence it is evident that there is an inherent requirement to find the best route for the lightpath, and after finding the route, assign wavelength to each link of the lightpath, so as to ensure WCC. The above is referred to as Routing and Wavelength Assignment (RWA) in optical networks. The RWA problem can be further subdivided into two sub parts- first the routing problem and second the wavelength

IV. GMPLS based Optical Networks

GMPLS brought in a common control plane standard for backbone networks. GMPLS employs dynamic provisioning of network services and Traffic Engineering (TE) [9] for prevailing backbone technologies. WDM as a technology promises enormous bandwidth potential, and hence to utilize its network resources efficiently, an intelligent control plane was an imminent requirement. GMPLS is the solution for an intelligent control plane of these networks. Further the basic concept of WDM lambda switching, i.e. to convert an input wavelength to an output wavelength was very similar to Multi Protocol Label Switching (MPLS) label mapping/switching [10]. Thus the concept of Multi Protocol Lambda Switching (MPLambdaS) was born for photonic networks. For optical networks using WDM technologies, MPLS techniques were used for establishing Label Switched Paths (LSPs) and further providing TE. This resulted in placing traffic on already computed and selected routes in the cloud or network, to optimize the utilization of network resources. This ensured routing service traffic away from congested subnets of the network. GMPLS employs Traffic Engineering Database (TED), which contains all information of the network, related to TE. In GMPLS a label associates to a specific resource, i.e. wavelength for a lambda switching cloud or network. Further GMPLS provides for bidirectional LSP [11] in contrast to unidirectional LSP in MPLS, which helps the control plane greatly. Another important facet of GMPLS is that the control and data

060.4256 Networks, network optimization

planes are separated, which is an inherent requirement for transport networks. The architecture for GMPLS was proposed and comprised of Path Computation Element (PCE)-based model [12]. The PCE can be described as an object or a component of a network node which can compute a route if it has been provided with a cloud or network graph. PCE can further apply very specific considerations or constraints during this mechanism of path estimation. It is designed to have a dedicated data plane to other PCEs for information sharing. A TED is defined as a matrix, which has all the information regarding the nature of cloud connectivity and all data regarding the available resources in that cloud. TED contains the topology and resource information of the domain. Further a Path Computational Client (PCC) is part of the architecture. PCC is described as a customer program application which actually requests the PCE for the specific path computation. The PCE then accesses the TED and other resource constraints, to optimally estimate the required path. The PCE can be located at any point in the domain of the network, but should be connected to all LSRs via the data plane. The protocol between PCE and PCC has been defined as PCE communication Protocol (PCEP) [13]. This specific send/receive communication protocol will contain the specific path request from PCC to PCE, and the estimated path calculated by PCE to PCC. The PCE after optimum estimation of the desired path will respond to PCC using the above communication channel, and in case of not being able to estimate the path, will intimate the same to the PCC with reasons for the path estimation failure. The performance of PCE-based routing with combined and separated routing schemes has been duly verified [14]. It has been seen that a central PCE for routing and wavelength assignment is an efficient architecture as shown in Fig. 1.

3
It has been observed in optical networks that service requests data rates are a miniscule fraction of the data rate that optical lightpaths generally provide. In DWDM technology, each lightpath may have a capacity of transporting up to 80 gigabits per second of optical traffic [2]. The user data rate requirements generally average out in the range 500 megabits per second. Hence we see that allocating a single service request to a lightpath in optical networks is a sheer waste of optical network resource. Hence, the need arises to map on more services on to a single lightpath, thus utilizing its capacity to the maximum. This in effect improves the capacity utilization of the optical network. The above phenomenon of optimum utilization of network resources in telecommunication networks had been termed as TE [15]. Traffic engineering is essential for Next Generation Networks (NGN) and is an ever growing field of study. Further optical networks require efficient RWA algorithms for protection against failures of optical links. The RWA combined with TE can provide maximum utilization of network resources. Though various studies and priority based algorithms for traffic grooming in optical networks have recently been undertaken [3] - [6], but the aspect of prioritization of user traffic to support Quality of Service (QoS), enforcing Service Level Agreements (SLAs) and use of fiber link state information has not been incorporated in these algorithms. The connection admission grooming algorithm proposed here, addresses all the above issues and is discussed in the following sections.

A. Dynamic Fiber State Information

In optical networks the present state of optical fiber links changes due to degradation in signal carrying capacity with time and impacts the effect of TE. Hence dynamic link state information is required, which will facilitate an efficient TE and optimum RWA. The architecture proposed for the above is the presence of Fiber Sensing Controllers (FSC) at each optical node or Label Switched Routers (LSR) of the network. The primary job of this sensing controller is to periodically measure the Optical Signal to Noise Ratio (OSNR), delay, jitter and Bit Error Rate (BER). These sensing controllers will compute the above parameters for every fiber link emanating from each LSR, and this information will be communicated to the central PCE controller. This information will be communicated on the PCEP towards the PCE. Further, at the PCE, all the information regarding the fiber link state will be analyzed and every fiber in the network will be classified as Fiber Class A (OSNR > 30 dbm), Fiber Class B (20 dbm < OSNR < 30 dbm) and Fiber Class C (OSNR < 20 dbm). This fiber link classification will be used in the CAG algorithm discussed in section E.

LSR Bidirectional Optical Links PCE Controller PCECP Link

LEGEND Fig. 1. PCE based Architecture

V. Connection Algorithm

Admission

Grooming

(CAG)

060.4256 Networks, network optimization

B. Class of Service Quantization of Traffic
The algorithm proposes to prioritize service traffic into various classes of service, so as to guarantee SLA. This prioritization will ensure that during congestion of subnet of the optical network, the lower priority service traffic is dropped and the higher priority service traffic is transported. To achieve the above, the algorithm classifies the service/user traffic into the following three classes:- Type A traffic (CoS = 1) for real time traffic (for voice and live video applications), Type B traffic (CoS = 2) for non real time traffic (for compressed video and transactional data traffic) and Type C traffic (CoS = 3) for delay tolerant traffic (transfer of files). The value for the CoS ranges from 3 to 1, as defined above. The highest value being 3 and the lowest value being 1. The SLAs with the user will be based on the above type of classification of traffic, and will provide the necessary Guarantee of Service (GoS). The above classification functionality will be performed by a traffic prioritizer called CoS Prioritizer, which will be present at each user/service interface in the network. The traffic requests as per SLAs, will be forwarded to the PCE for further processing, using the algorithm discussed in section E.
USER TFC

LSR with PCE component Bidirectional Optical Links CoS Quantiser PCE Controller Routing Controller

Control Links LSP Prioritizer Fiber Sensing Controllers Signalling Controller PCEP Link

LEGEND
Fig. 2. Architecture for CAG Algorithm

E. Proposed Algorithm Functionality

The various parameters available with the PCE are as follows: (a) Type of fiber (A/B/C) received from all the FSCs in the network. (b) Class of traffic (A/B/C) received from the CoS Quantiser, at each servicer/user access interface. (c) Number of hops of LSP requested received from LSP Prioritizer. (d) Data rate of LSP requested received from LSP Prioritizer. The various parameters to be used in the algorithm are:(a) Mh : The maximum number of hops from any source to any destination in the network, as calculated by the routing algorithm Resource Reservation Protocol Traffic Engineering (RSVP-TE) [16], which will be used in the proposed algorithm. (b) Md: The maximum data rate configured for each link, assuming that all links in the network have equal bandwidth. The data rate requirement of the (c) DLSP: individual LSP request. (d) WLSP: Weight of each individual LSP (to be defined in step 2). (e) HLSP: Number of hops required by the individual LSP (f) CLSP: Class of service of user traffic, carried by the individual LSP. The above parameters are available with the PCE, and are processed according to the following steps:

C. Prioritization of Label Switched Paths

Any service request in the network, from a certain source to a certain destination, is translated into a unique LSP. In the GMPLS architecture the PCC will render a connection requirement to the PCE for path computation of every LSP. The PCE will then involve an efficient routing algorithm for establishment of the LSP. A LSP prioritizer will be employed for the proposed CAG algorithm. This LSP prioritizer will extract two important parameters from the LSP request; firstly, the number of hops required by the LSP (using the dynamic routing information already available with the PCE) and secondly, the data rate required by the LSP. These two parameters will be utilized in the CAG algorithm discussed in section E.

D. Proposed Algorithm Architecture

The various components of the proposed algorithm architecture are: Fiber Sensing Controllers, CoS Quantizer, LSP Prioritizer, central PCE Controller, central Routing Controller, Signaling Controllers, LSRs with PCE components, bidirectional optical links, control links and PCECP links. The detailed architecture for the above components for implementation of the CAG algorithm is shown in Fig. 2.

060.4256 Networks, network optimization

Step 1: Create a LSP Requirement Matrix for the LSP requests in the network, as shown in Table 1.a.
Ser No 1 2 3 4 5 6 Sourc e Id 01 02 01 02 03 05 Dest Id 03 05 04 06 06 02 No of Hops (HLSP) 4 3 2 4 3 3 Data Rate (in Gbps) (DLSP) 0.5 1.0 0.5 1.5 2.5 1.5 CoS (CLSP) 2 1 3 2 3 1 1 2 3 4 5 6 Ser No Sou -rce Id 01 02 01 02 03 05 Dest Id 03 05 04 06 06 02 No of Hops (HLSP) 4 3 2 4 3 3 Data Rate (in Gbps) (DLSP) 0.5 1.0 0.5 1.5 2.5 1.5 CoS reqd (CLSP) 2 1 3 2 3 1

5
Weight (WLSP) 1.716 1.183 1.550 1.816 2.000 1.234

Table 1.a: LSP Requirement Matrix (withPCE controller)

Table 1.b: LSP Weight Matrix (with PCE controller)

Note: The values given in the matrix are used to demonstrate the functionality of the algorithm. Step 2: After obtaining the above LSP requirement matrix from all LSRs, the PCE will calculate the weight (WLSP) for each LSP by using the equation 1. WLSP = { HLSP / ( Mh - HLSP +1) } + { CLSP / (3 1) } + { DLSP / ( Md - DLSP + 1 ) }

Note: Mh has been chosen to be 4 hops and Md has been chosen as 10 gigabits per second, to demonstrate the functionality of the algorithm. Step 3: The entries in LSP Requirement Matrix at Table 1.b will be rearranged in ascending order of the weights (WLSP) of LSP such that, light weighted LSPs are higher on the table and heavier LSPs are at the bottom of the table. Resulting, that a Prioritized Queuing Table will be created as shown in Table 2.
Ser No 2 6 3 1 4 5 Sou -rce Id 02 05 01 01 02 03 Dest Id 05 02 04 03 06 06 No of Hops (HLSP) 3 3 2 4 4 3 Data Rate (in Gbps) (DLSP) 1.0 1.5 0.5 0.5 1.5 2.5 CoS reqd (CLSP) 1 1 3 2 2 3 Weight (ascendi -ng) (WLSP) 1.183 1.234 1.550 1.716 1.816 2.000

CLSP + - (1)

The equation 1 has been devised with the concept that the weight of an LSP request (WLSP) is directly proportional to the fraction of network resources used by the LSP, assuming that the request is serviced. This capacity utilization factor, which is coined as weight of the LSP request (WLSP), is calculated with the summation of three terms in equation 5.1. The explanation of the three terms is as given below: (a) First term of equation 1 gives the ratio of number of hops used (HLSP), with respect to the maximum number of hops in the network (Mh). (b) Second term of equation 1 gives the ratio of value of CoS (CLSP) assigned, with respect to the highest value of CoS, which is 3 as explained in section 5.3. (c) Third term of equation 1 - Fraction of data the rate utilized (DLSP), with respect to maximum data rate engineered in the network (Md), which has been assumed for the algorithm to be 10 Gigabits per second. The WLSP values for all the entries in the LSP requirement matrix in Table 1.a will be calculated, and a new LSP weight matrix will be created as shown called as Table 1.b

Table 2: Prioritized Queuing Table (with PCE controller)

Step 4: The fiber architecture of the network for this algorithm is as shown in Figure 3. From the architecture of Figure 3, node 1 is connected to node 2, node 3 and node 4 with 12 fiber links each, resulting in a total of 36 fiber links originating from Node1.
Note:- All optical connections are 12 fiber links each . LSR with PCE component Bidirectional Optical Links Fiber Sensing Controllers

3 1 5 4

LEGEND

Fig. 3: Fiber Architecture of the Network.

060.4256 Networks, network optimization

From the dynamic fiber state information received by the PCE controller, the PCE will create Nodewise Fiber State Information for each node. The node wise fiber state information for node 1 is given at Table 3.
Destinati on Node 2 3 4 Number of Fibers (Class A) 03 02 04 Number of Fibers (Class B) 07 06 05 No of Fibers (Class C) 02 04 03

6
(b) Each bi-directional link has a maximum data rate capacity of 10 gigabits per second. The minimum data rate unit for a LSP request has been kept at 500 megabits per second. Further LSP requests can be multiples of 500 megabits per second. (c) The packets are generated randomly among all possible source destination pairs. (d) A request for connection is said to be not serviced or rejected if it is not dispensed with the available network resources. (e) The achievement of the designed CAG algorithm is validated in terms of Blocking Probability (BP), wherein; the inferior the BP, the better the performance of the network. The simulation has been carried out in OMNET++ (version 4.2.2), which is a discrete event simulating environment. INET framework has been used inside the OMNET++ environment, which facilitates simulation of optical networks. Optical routers incorporating RSVP-TE have been used in the INET framework. The simulation has been carried out in four phases as described below: Phase I In the first phase of the simulation process, the functionality and performance of CAG algorithm was evaluated in isolation, without incorporating any routing algorithm. The simulation was initially carried out with one wavelength per link in the algorithm architecture as given in Figure 2. The simulation was carried out in two parts. In the first part, the network was simulated with ordinary traffic (without any traffic grooming /incorporating traffic engineering algorithm). In the second part, the simulation in the network has been carried out incorporating the CAG algorithm. The blocking probability (BP) has been plotted with respect to increasing Traffic Intensity (TI), both in logarithmic scales for single wavelength, and the result is shown in figure 4(a).
Links with 1 wavelength

Table 3: Nodewise Fiber State Information (for Node 1).

Step 5: The Prioritized Queuing Table (for each node), Table 2 from Step 3 and Nodewise Fiber State Information, Table 3 from Step 4, will be used by the PCE controller for assignment of LSPs for each segment of the entire path. The top entry of Table 2, will be assigned a higher class of fiber from Table 3 for the destination node, if available; else the next class of fiber which is available in that segment will be allotted. If no fibers are available, the LSP request will be blocked. This process will be repeated for the next entry of Table 2, till there are no entries left in Table 2. Step 5 will be repeated for all nodes in the network architecture.

F. Results and Discussion

The network architecture which has been assumed for simulation of the proposed CAG algorithm using dynamic RWA in PCE based architecture consists of 11 LSRs (optical nodes) and 14 bidirectional optical links. Each bidirectional link comprises of 4 wavelengths each. The architecture has one central PCE controller connected to each LSR by a PCECP link for transmission of control signals from PCE to respective LSRs and vice versa. Each LSR is provided with a FSC, which regularly updates the fiber state information to the central PCE controller via the PCECP link. A single routing controller is co-located with the central PCE controller and is responsible for routing of LSPs in this domain. The routing controller uses GMPLS RSVP-TE protocol to find the optimum routes for each LSP. A signaling controller is collocated with each LSR. The signaling controller communicates with other signaling controllers using the control link. GMPLS signaling protocol is used to establish the LSP. The following assumptions have been made for the simulation process:(a) The connection requests are random in nature and are established on a process which is Poisson in nature. The times of arrival between two events of connection requests in succession, follow a distribution which is exponential in nature.

P r o b

B lo c k in g P ro b a lit y

B l o c k i n g

Groomed Tfc Ordinary Tfc

10

10

-1

10

-0.2

10

-0.1

Traffic Intensity

Traffic Intensity

Fig. 4 (a). Performance of CAG algorithm, BP v/s TI for single wavelength.

060.4256 Networks, network optimization

The simulation was next carried out with two, three and four wavelengths in each of the bidirectional optical links. The simulation plots of BP with respect to TI for two, three and four wavelength links for both ordinary traffic and groomed traffic (with CAG algorithm) is shown in figures 4 (b) to 4 (d) respectively.
Links with 2 wavelengths

7
From the simulation results of figure 4(a) to 4(d), we can infer: (a) The plot shifts to the right as the number of wavelengths used with each link increases, thus showing that the network experiences lesser connection BP as the number of wavelengths increases on every link. (b) There is a substantial improvement in the connection BP performance of the network while incorporating CAG algorithm. (c) The improvement due to the CAG algorithm is most effective when a single wavelength is used in each link. The improvement reduces as the number of wavelengths increase in each link. This is in accordance to the established fact, that TE is less effective when the network is underutilized. Phase II In the second phase of simulation process, the CAG algorithm has been combined with dynamic RSVP-TE routing model and its performance evaluated by comparing it with a dynamic RSVP-TE routing model without CAG algorithm The blocking probability (BP) has been plotted with respect to increasing Traffic Intensity (TI) for four wavelengths, both in linear scales, and the result is shown in figure 5.
Comparision of Dynamic Routing with and without CAG algorithm 0.7

B 10 l o c k i n 10 g
B lo c k in g P ro b a lit y

Ordinary Traffic Groomed Traffic

-1

P r o b 10

-2

10

0.1

Traffic Intensity Traffic Intensity

10

0.2

Fig. 4 (b): Performance of CAG algorithm, BP v/s TI for two wavelengths.

Links with 3 wavelengths Ordinary Traffic Groomed Traffic

B 10 l o c k i n 10 g
B lo c k in g P ro b a lit y

-1

B lo c k in g P ro b a b ilit y

P r o b 10

-2

B l o c k i n g
10
0.3

0.6

dynamic wo CAG dynamic with CAG

0.5

0.4

0.3

Traffic Intensity Traffic Intensity

10

0.4

Fig. 4(c): Performance of CAG algorithm, BP v/s TI for three

Links with 4 wavelengths

P r o b

0.2

0.1

B lo c k in g P ro b a lit y

B l 10 o c k i n g 10 P r o b
10

Ordinary Traffic Groomed Traffic

0 2.4

2.6

2.8

3 Traffic Intensity

3.2

3.4

3.6

3.8

Traffic Intensity Fig 5: Comparison of Dynamic Routing with and without CAG algorithm, BP v/s TI for four wavelengths.

-1

-2

10

0.4

Traffic Intensity Fig. 4 (d): Performance of CAG algorithm, BP v/s TI for four wavelengths.

10 Traffic Intensity

0.5

From the simulation result of figure 5, we can infer: (a) Dynamic routing (RSVP-TE) incorporating CAG algorithm has a better performance in terms of connection BP. There is a substantial improvement in connection BP using CAG algorithm. (b) The improvement in connection BP is most effective when the network is loaded with greater than 90% of input user traffic. Phase III process, the In the third phase of simulation dynamic routed (RSVP-TE) CAG

060.4256 Networks, network optimization

algorithm was combined with two of the wavelength assignment techniques. In the first part of simulation Random Fit (RF) wavelength assignment technique was used and in the second part First Fit (FF) wavelength assignment technique was used. The simulation result comparing the performance of dynamic routed (RSVP-TE) CAG algorithm combined with RF and FF wavelength assignment techniques is shown in figure 6. The connection BP has been plotted with respect to increasing Traffic Intensity (TI) for four wavelengths, both in linear scales.
Comparison of RF and FF WA schemes

8
scales.
Comparison of CAG and Priority Algorithm 5

4.5

B lo c k in g P ro b a b ilit y

B l o c k i n g P r o b

CAG Algorithm Priority Algorithm

3.5

2.5

1.5

B lo c k in g P ro b a b ilit y

B l o c k i n g P r o b

0.6

Random Fit First Fit

0.5

0 3.65
0.5

3.7

3.75

Traffic Intensity Fig 7: Comparison of CAG and Priority Based Algorithm, BP v/s TI for four wavelengths.

3.8 Traffic Intensity

3.85

3.9

3.95

0.4

0.3

0.2

0.1

0 3.7

3.75

3.8 Traffic Intensity

3.85

3.9

3.95

Traffic Intensity Fig 6: Comparison of Dynamic Routing with CAG algorithm sing RF and FF WA schemes, BP v/s TI for four wavelengths.

From the simulation result of figure 6, we can infer: (a) FF WA scheme has lower connection BP than RF WA scheme at higher input traffic load in the network. Hence we see that for the same value of TI, BP is lower for FF WA when compared with RF WA scheme. (b) This is due to the fact that RF WA scheme increases discontinuous wavelength fragmentation, and thus degrades the forward blocking probability in succeeding hops, during WA under WCC.
Phase IV In the fourth and last phase of simulation process the performance of CAG algorithm combined with dynamic routing has been compared with Priority based routing algorithm [10] and the simulation result is shown in figure 7. The connection BP has been plotted with respect to increasing Traffic Intensity (TI) for four wavelengths, both in linear

From the simulation result of figure 7, we can infer: (a) The proposed CAG algorithm performs substantially better than the Priority based algorithm. (b) This is due to the fact that the proposed CAG algorithm uses four variables with respect to LSP requests viz. CoS, number of hops, data rate and fiber state to calculate priority for each lightpath request. In comparison, the Priority based algorithm uses only two variables viz. number of hops and data rate of requested LSP, to accord priority to each lightpath request. (c) Further, for the data rate variable criterion, the Priority based algorithm accords higher priority to lightpath requests which demand greater volumes of data rate, resulting in that fewer number of lightpath requests are serviced in the network. (d) The CAG algorithm on the other hand gives priority to lightpath requests, which require lower volumes of data rate and effectively can service greater number of lightpath requests. This fact is evident on observing their comparison in terms of BP in figure 7. Next both the dynamic routed CAG and Priority algorithm have been combined with two of the wavelength assignment techniques. Firstly, dynamic routed CAG and Priority algorithm has been combined with RF WA scheme.

060.4256 Networks, network optimization

The performance of dynamic routed CAG and Priority algorithm combined with RF WA scheme has been simulated and its comparison demonstrated in the simulation result shown in figure 8. The blocking probability (BP) has been plotted with respect to increasing Traffic Intensity (TI) for four wavelengths, both in linear scales.
Comparison of CAG and Priority Algo Random Fit 0.4

Comparison of CAG and Priority based algo combined with FF WA 0.25 CAG algo with FF Priority based algo with FF

B lo c k in g P ro b a b ilit y

B l 0.35 o c 0.3 k i 0.25 n g 0.2 P 0.15 r o b 0.1

0.05 0

Prioritybased aggo with RF CAG algo with RF

P 0.1 r o b
0.05

B lo c k in g P ro b a b ilit y

B l o 0.2 c k i 0.15 n g

0 1.84

1.86

1.88

1.9 Traffic Intensity

1.92

1.94

1.96

1.98

Traffic Intensity Fig 9: Comparison of RF WA schemes combined with CAG and Priority Based Algorithm, BP v/s TI for two wavelengths.
3.76 3.78 3.8 3.82 3.84

Traffic Intensity Fig 8: Comparison of FF WA schemes combined with CAG and Priority Based Algorithm, BP v/s TI for two wavelengths.

3.86 Tfc Intensity

3.88

3.9

3.92

3.94

3.96

From the simulation result of figure 8, we can infer: (a) RF WA scheme combined with CAG algorithm performs better than RF WA scheme combined with Priority based algorithm when the input traffic load in the network is more than 80 percent. (b) Higher the input traffic load in the network, more scenarios of congestion in the network and hence the better algorithm, which is inferred as the CAG algorithm in figure 8, observes a lower BP. (c) Hence the performance of CAG algorithm is better than Priority based algorithm when combined with RF WA scheme. Finally, dynamic routed CAG and Priority algorithm have been combined with FF WA scheme. The performance of dynamic routed CAG and Priority algorithm combined with FF WA scheme has been simulated and its comparison is demonstrated in the simulation result shown in figure 9. The blocking probability (BP) has been plotted with respect to increasing Traffic Intensity (TI) for two wavelengths, both in linear scales.

From the simulation result of figure 9, we can infer: (a) FF WA scheme combined with CAG algorithm performs better than FF WA scheme combined with Priority based algorithm when the input traffic load in the network is more than 88 percent. (b) Higher the input traffic load in the network, more scenarios of congestion in the network and hence the better algorithm, which is inferred as the CAG algorithm in figure 9, observes a lower BP. (c) ence the performance of CAG algorithm is better than Priority based algorithm when combined with RF WA scheme. The results and their analysis as shown above, demonstrate that incorporating the proposed CAG algorithm with combined RWA in PCE based architecture, the connection blocking capability, which is an important parameter for judging the efficiency of any network, reduces substantially. This in turn increases the utilization of network resources substantially. Consequently the proposed CAG algorithm improves the efficiency of pure optical networks. The CAG algorithm has better connection blocking performance than any proposed connection grooming algorithm to the best of our knowledge.

060.4256 Networks, network optimization

G. Conclusion
Intelligent optical networks (GMPLS with intelligent control plane) are the future of backbone telecommunication networks. These optical networks are the important backbone interconnecting constituents, of various emerging high data rate wireless networks, which are seen as future of mobility networks. But with more and more end users hogging the backbone bandwidth, the massive bandwidth offered by optical networks may soon become inadequate given the increasing trends of user traffic in various access networks. CAG algorithm combined with RWA in PCE based architecture incorporates connection admission grooming and traffic engineering to bring in an efficient methodology to maximize the capacity utilization of optical networks. The results and their analysis thereafter, demonstrate that incorporating the proposed CAG algorithm with combined RWA in PCE based architecture; the connection blocking capability reduces substantially. This in turn increases the utilization of network resources and consequently improves the efficiency of pure optical networks. The CAG algorithm has better connection blocking performance than any proposed connection grooming algorithm to the best of our knowledge. [8]

10
Optical Networks, IEEE Journal on selected areas in Communication, Vol. 14, No. 5, pp.

[9] [10] [11] [12] [13]

[14]

REFERENCES
[1] [2] [3] B. Mukherjee, Optical Communication Networks, Mc Graw-Hill, New York, 1997. R. Ramaswami and K. N. Sivarajan, Optical Networks, Elsevier, 2004.

[15]

[16]

[4]

[5]

K. Zhu, B. Mukherjee, A Review of Traffic Grooming in WDM Optical Networks: architectures and challenges, Optical Networks Magazine, Mar/Apr. 2003. P. Iovanna, M. Settembre, R. Sabella, G. Conte and L. Valentini, Performance Analysis of a Traffic Engineering Solution for Multilayer Networks Based on the GMPLS Paradigm, IEEE Journal on Selected Areas in Communication, vol. 22, no.9, pp. 17311740, Nov. 2004. T. Deng and S. Subramaniam, Adaptive QoS Routing in Dynamic Wavelength- Routed Optical Networks, 2nd International IEEE B. C. Chatterjee, N. Sarma and P. P. Sahu, Priority Based Routing and Wavelength Assignment with Traffic Grooming for Optical Networks, J. Opt. Commun. Netw., vol. 4, no 6, pp. 480-489, Jun. 2012. D. Bannerjee and B. Mukherjee, A Practical Approach for Routing and Wavelength Assignment in Large Wavelength-Routed

903-908, Jun. 1996. H. Zang, J. P. Jue, and B. Mukherjee, A Review of Routing and Wavelength Assignment approaches for WavelengthOpt. Routed Optical WDM Networks, Networks Mag., Vol. 1, No. 1, pp. 47-60, Jan. 2000. A Farrel and I Bryskin, GMPLS Architecture and Applications, Elsevier, 2006. U Black, MPLS and Label Switching Networks, 2nd Edition, Prentice Hall, 2002. L. Berger, Generalized Multi-Protocol Label Switching (GMPLS) Signalling Functional Description, IETF RFC 3471, Jan. 2003. A Farell, J. P. Vasseur and J. Ash, A Path Computation Element (PCE)-based architecture, IETF RFC 4655, Aug. 2006. J. Ash and J. L. Le Roux, Path Computation Element (PCE) communication protocol generic requirements, IETF RFC 4657, Sep. 2006. Y. Zhao, J. Zhang, Y. Ji and W. Gu, Routing and Wavelength Assignment Problem in PCE - Based Wavelength Switched Optical Networks, J. Opt. Commun. Netw., Vol. 2, No. 4, pp. 196-205, Apr. 2010. D. J. Malcolm, B. Agogbua, M. ODell and J. Mcmanus, Requirements for Traffic Engineering over MPLS, IETF RFC 2702, Sep. 1999. L. Berger, Generalized Multi-Protocol Label Switching (GMPLS) Signalling Resource Reservation Protocol-Traffic Engineering (RSVP-TE) Extensions, IETF RFC 3473, Jan. 2003.

Sidharth Shukla- The author completed M Tech (Comm Sys) from Indian Institute of Technology, Roorkee. He is presently pursuing Phd at Indian Institute of Technology, Indore. Vimal Bhatia- The co-author is presently Associate Professor in Electical Engineering Dept. at Indian Institute of Technology, Indore. He has a mix of academic and industrial experience both in India and the UK. He completed Ph.D. from Institute for Digital Communications at The University of Edinburgh (UoE), UK. He received M. Tech. in Signal Processing from NSIT, Delhi in 2001. He has various publications in renowned Journals and conferences. He is a reviewer for IEEE, Elsevier, and IET. He is currently Senior Member of IEEE and certified SCRUM Master.

[6]

Conference on Broadband Networks, Broadnet, vol.1, pp. 184-193, 2005.

[7]