simulnet: a wavelength-routed optical network simulation...

SimulNet: a Wavelength-Routed Optical Network Simulation Framework

Francesco Palmieri, Ugo Fiore and Sergio Ricciardi*Federico II University ofNapoli

Via Cinthia, 5, Complesso Universitario di Monte S. Angelo, 80126 Napoli, Italy{francesco.palmieri, ugo.fiore, sergio. ricciardi} @unina. it

Abstract

Simulation seems to be the best available alternativeto the deployment of expensive and complex testbedinfrastructures for the activities oftesting, validating andevaluating optical network control protocols andalgorithms. In this paper we present SimulNet, aspecialized optical network simulation environmentproviding the foundation for the study and analysis ofthekey control plane characteristics of wavelength-routednetworks. Such an environment would provideresearchers with an open framework for easily exploringthe evolving characteristics of WDM-routed technologieswhich includes developing new protocol suites orperforming rapid evaluation and easier comparison ofresults across research efforts.

Keywords: simulation, WDM-switching, RWA, axe

1. Introduction

Simulation plays an important role in networkprotocol design, providing researchers with a costeffective method to analyze and study the behavior ofproposed protocol models. Particularly, simulationbecomes an indispensable tool in WDM-basedtransparent optical networking research, since it helpsresearchers to quickly and inexpensively validate andevaluate the performance of new protocols andalgorithms without the need of installing and managingcomplex network testbeds that require very expensiveoptical devices and communication infrastructures. Infact, even on a fully featured testbed, not all algorithms,mechanisms and protocols can be readily implemented,tested and evaluated, because most of the availableoptical network equipments are proprietary vendorproducts featuring a very low degree ofprogrammability.Implementing new control plane features and algorithmson top of such proprietary equipment can be a verytedious and difficult task, since it requires accessing,controlling and programming the low level capabilities ofthese devices. Unfortunately only a very limited numberof simulation tools are available for conducting research

related to different routing and wavelength assignmentalgorithms (RWA), topology management (e.g. converterplacement algorithms), re-optimization, centralized anddistributed wavelength reservation schemes, and theeffect on RWA algorithms of inaccuracies in networkwide state information or physical-level impairments.Furthermore, most of the available solutions are limitedin scope, difficult to use or not totally open. In manycases, they have also been based on simulation modelsdesigned specifically to cope only with a specificproblem. Since each of these solutions used their ownsimulation platform, model and assumptions, it isgenerally difficult to reuse existing protocol modules andcompare simulation results under a common assessmentenvironment. Our aim is to address these problems withthe introduction of a new simulation frameworkdeveloped specifically for the testing and performanceanalysis of distributed dynamic RWA algorithms onoptical networks and also for optimization algorithms andprotocols validation. The framework we present, calledSimulNet, has been designed to be easy to use andexhibits satisfactory performance also when simulatinglarge network topologies. It has been implementedaccording to a modularized, platform-independent, andextensible architecture and provides a useful baselinelibrary ofcommonly used RWA algorithms and signalingschemes. We based our framework on a totally flexiblenetwork model, supporting heterogeneous WDMequipment, with or without wavelength conversioncapability, in which the number and type of lambdas canvary on each link. It provides a fully dynamic andconfigurable path selection scheme supporting subwavelength bandwidth allocation (grooming).Furthermore, it allows simulations to be aware of all thecomplexity, expensiveness, performance and resourcelimitation constrains implicit in the various flavors ofoptical switching devices, for example explicitly andproportionally penalizing, when instructed to do so, allthe paths that require wavelength conversion. Thesimulator has been successfully validated byimplementing some well-known algorithms, andcomparing the results with those available in literature.

* Sergio Ricciardi is PhD student at the Department ofComputer Architecture ofthe Technical University ofCatalonia (UPC), JordiGirona 3, 08034 Barcelona, Catalunya, Spain.

978-1-4244-4671-1/09/$25.00 ©20091EEE 281

Authorized licensed use limited to: Universita degli Studi di Napoli. Downloaded on August 31, 2009 at 17:46 from IEEE Xplore. Restrictions apply.

2. Related work

At the state of the art, several simulation packages andtools modeling optical network infrastructures areavailable, but none of them totally provides the necessarysupport and flexibility needed for the study and theperformance assessment of new routing, signaling andwavelength management schemes on modem WDMempowered networks. GLASS (GMPLS LightwaveAgile Switching Simulator) [1], developed at NIST, is anoptical network simulation tool built on the SSFNETFramework [2]. It supports many advanced features,including GMPLS and QoS, and several optical failurerecovery schemes. Since GLASS is implemented in Javait is easy to configure, modular and platformindependent. Unfortunately, its complexity adverselyconditioned its performance and hindered thedevelopment of new routing schemes on this platform.OWNS [3] is tool that is being used for research in theoptical network domain. It is built on the well knownmature foundation of the NS-2 [4] simulationinfrastructure with a set of additional optical WDMextensions. Unfortunately, with these benefits it alsoinherited some inconveniences of NS-2, such as bloatedcode base, slow execution speed and large memoryfootprint. Also, configuring and implementing any newrouting scheme or protocol requires extensive and carefulcoding, because of the absence of a modular structure.OWNS also lacks the feature set required for theimplementation of converter placement algorithms.JAVOBS [5] is a set of flexible and configurable javalibraries dedicated to the simulation of Optical BurstSwitched (OBS) networks. It supports simulation ofvarious OBS schemes, reservation protocols andscheduling algorithms, on any topology of reasonablesize. SimulNet can be considered quite similar toJAVOBS as for the general simulation framework andthe configurability, with the fundamental difference thatwhile JAVOBS operates over OBS networks, SimulNetis expressly designed to operate over wavelengthswitched networks' control plane layer.

3. Basic Architecture

Simulation environments allow the user to predict thebehavior of a set of network devices on a complexnetwork, by using an internal model that is specific to thesimulator. Simulators do not necessarily reproduce thesame sequence of events that would take place in the realsystem, but rather apply an internal set of transformationroutines that brings the simulated network to a final statethat is as close as possible to the one the real systemwould evolve to. To accomplish this task, simulatorswork on a model that contains all the relevantinformation that must be observed, while abstracting

978-1-4244-4671-1/09/$25.00 ©2009 IEEE

irrelevant details, thus simplifying both the simulationand the analysis of the network. This approach typicallyallows the simulated network to scale well in size andcomplexity, as irrelevant details may be totally abstractedor represented in a simplified manner. Thus, thesimulator can be used to manage also complex networkswith hundreds or thousands of network elements andwavelengths, while greatly reducing the probability ofintroducing programming bugs. The drawback is that thesimulated devices may have limited functionalities andtheir behavior may not closely resemble that of realworld devices. This is the reason why both theparameters to be represented or abstracted in the modeland the representation detail must be carefully chosen.An excessive degree of sophistication in the simulationenvironment is often the cause of a limited flexibility inconfiguration and introduction of new functionalmodules, and in general heavily taxes both the runtimeperformance and the ease of use. In fact, the realobjective of a simulation is to make available asustainable and effective model of the involved systemthat is sufficiently accurate to reliably analyze andevaluate only the specific set of properties of the systemthat are under observation. Furthermore, simulationresults are easier to analyze than experimental onesbecause important information at critical points can beeasily logged to help researchers in diagnosing thenetwork behavior. Starting from these premises, theproposed simulator architecture has been designedaccording to a fully modular scheme, to accommodatemost of the control-plane specific characteristics ofwavelength-routed network and provides a useful set ofnetwork traffic generators. The simulator models opticalfibers and wavelengths individually for each link, thusallowing maximum flexibility in the representation ofdissimilar networks. It also provides network utilitysubroutines to configure, monitor, and gather results andstatistics about simulated networks. The resultingframework provides great configuration flexibility intopology definition (each node can or cannot supportconversion capability; the number and type of lambdas,the associated maximum bandwidth and propagationdelay can be specific for each link) and enables easyextension to introduce new features, not only in terms ofRWA and grooming algorithms but also allowing otheraspects of the simulator to be customized or expanded ifdesired in the future. The whole framework has beendesigned as an object-oriented application in which allfunctional entities are implemented by individual objectsor modular entities interacting with each other. Eachobject can contain or refer to other objects for thecomposite identification of an operation. Objects can alsobe abstracted and encapsulated to facilitate extension. Forexample, if a new optical node type or new RWAalgorithm in the simulator is needed, it can be

282


conveniently added by extending the existing networkdevice object or RWA algorithm object. The keycomponents of the simulator can be divided into physicallayer abstractions, such as optical switching devices andmulti-wavelength fiber links, and logical layer modulesworking upon them, such as the RWA simulation engineand the traffic generation module, which together createand maintain the virtual topology.

3.1. Modeling the physical layer

The OXC object models the various types ofswitching nodes that constitute the Transport Plane of thesimulated optical network. OXCs are responsible forswitching traffic from an incoming fiber/wavelength pair(or electrical link) to an outgoing one. The necessarymapping information is maintained in a special internaltable, called the switching matrix, which the OXC looksup before accomplishing each switching operation.According to their routing, traffic grooming andwavelength conversion capabilities of their interfaces,OXC objects can be used to model both hybrid opticalrouters, add and drop multiplexers or pure wavelengthrouters, operating in transparent or opaque mode. Theyprovide wavelength routing & switching, wavelengthconversion, fiber/port switching, waveband switchingand regeneration. The simulated optical switchingfunctions are supported in either transparent opticalswitching architecture or in the opaque one with O-E-Oconversions. According to the underlying architecture thelambda conversion/switching capabilities may havedifferent limitations, such as the number of converters inan OXC and the range of conversion. Each fiber featuresan independent WDM capability, i.e. consists of severalwavelengths and the actual number of wavelengths is aparameter that can be configured for each fiber.Moreover, within each fiber strand connecting a coupleof OXC objects there is one additional virtual linkrepresenting the dedicated communication channelneeded to convey control plane traffic.

3.2. The RWA engine

The RWA simulation engine that is the heart of ourframework is composed of two components: routingmodule and wavelength assignment module.

3.2.1. Routing. The routing module supports thedynamic creation and deletion of unidirectionallightpaths by determining the routes needed to establishsuch paths in the current network topology and with theavailable resources. It maintains two distinct views of thenetwork - the physical and the virtual topology represented as multi-graphs in which each edgecorresponds to an individual channel (wavelengths), each

978-1-4244-4671-1/09/$25.00 ©2009 IEEE

fiber can support multiple channels, and there can bemore than one fiber connecting the same pair ofnodes.

The physical topology represents the real networkinfrastructure where every link between adjacentswitching nodes and its resources (wavelengths andbandwidths) and conversion capabilities are catalogued.The virtual topology represents the current networkstatus, dynamically modified during the creation oflightpaths. Multiplexing and demultiplexing wavelengthsis simulated implicitly in the virtual topology forwardingplane. The simulator switching logic within the RWAengine will ideally bridge the incoming wavelength andoutgoing wavelength to set up lightpaths. In aconfigurable manner, lightpath set-up is realized eitherfirst on existing available lightpaths (according to a bestfit strategy in case there are several lightpaths satisfyingthe connection request), or directly executing the RWAalgorithm without consulting the lightpath allocationtables. In the current implementation, the RWA enginealready provides some basic predefined algorithms, suchas Dijkstra shortest path algorithm/Minimum HopAlgorithm (MHA), Shortest Widest Path Algorithm(SWP), MINimum Residual Capacity algorithm(MinRC), k-shortest path algorithm and MinimumInterference Routing Algorithm (MIRA) [6][7][8].

3.2.2. Wavelength Assignment. The wavelengthassignment module is responsible for selecting thewavelengths needed for lightpath creation. Our simulatorsupports optional wavelength conversion capability oneach OXC. Also, channel allocation is done separatelyfor each link. Up to 64 wavelengths can be used fortransmission between network elements. More than onewavelength channel can be assigned between twonetwork elements depending on traffic demands.Wavelength reuse is employed when possible. Most ofthe predefined algorithms of our RWA engine treatsrouting and wavelength assignment as a single jointproblem, that is they search paths for connection requestsconsidering both the routing and the wavelengthassignment activities at the same time. Nevertheless, oursimulator supports also disjoint RWA, in which theabove activities are managed as two distinct problems.First the routing algorithm finds a suitable path for theconnection in terms of available fibers between sourceand destination nodes; then, the wavelength assignmentprocedure provides wavelength assignments according toa specific criterion aimed at optimizing some resources.Different wavelength assignment algorithms have beenimplemented: frrst-fit, most-used/pack, least-used/spread,random allocation, round robin. With the first-fitalgorithm, the first available wavelength is chosen,causing the wavelength with the smallest index to bechosen more often. The most-used/pack algorithm selects

283


the available wavelength which is currently utilized onthe largest number of fibers, thus preserving themaximum number of different wavelengths for futurerequests that may have the wavelength continuityconstraint; analogously, the least-used/spread algorithmselects the available wavelength which is currentlyutilized on the smallest number of fibers, thus trying tobalance the load on the network. The random ruledistributes the connection traffic randomly so thataverage wavelength utilizations are balanced. In theround robin allocation scheme wavelengths are indexedand assigned in a circular manner, thus maximizing theuse of all the available wavelengths in the network inorder to minimize the blocking probability due towavelengths unavailability. Finally, the bandwidth on awavelength can be divided into smaller sub-ratecapacities called sub-wavelength units, to be assigned tospecific end-to-end connections. A connection requestcan demand one or more sub-wavelengths to a maximumof the whole available bandwidth supported on a specificwavelength, by performing traffic grooming fromproperly capable devices. Our simulation environmentsupports both single-hop and multi-hop groomingcapability [9]. Single-hop grooming will be only possibleon end-to-end connections realized on a singlewavelength. On the other side, multi-hop grooming offersmaximum flexibility in allowing sub-wavelengthallocation across multiple chained lightpaths or singlelightpaths built on multiple wavelengths (when thecontinuity constraint is relaxed). Clearly our networkmodel allows the RWA algorithms to be aware of thedifferent cost and complexity of the two mechanisms.

3.2.3. Signaling. The simulated control plane signalinglogic is based on a reverse path reservation model thatcan be viewed as a simplified subset of RSVP-TE. Theconnection setup procedure is divided into two phases:downstream and upstream. During the downstreamphase, each of the OXC nodes determines whether therequired destination is reachable and there are enoughresources to accommodate the new connection. As soonas the RWA engine determines on each node both theoutgoing fiber and a wavelength with adequatebandwidth resources, it temporarily allocates them byadding a new record on the switching matrix for the newconnection. At this instant, the setup is still incomplete,and therefore the switching matrix record is flagged asprovisionally reserved and cannot yet be used by theOXC. At each step, the involved node passes the requestto the downstream OXC object. If there is nofiber/wavelength acceptable for routing the currentrequest or the available bandwidth is inadequate, thenode rejects the requests and sends a response back to thesource. The upstream phase starts as soon as the requestreaches the egress node of the connection that

978-1-4244-4671-1/09/$25.00 ©2009 IEEE

acknowledges it along the same path. During this phase,when any node receives an acknowledge from itsdownstream node, it permanently allocates the resourcesreserved during the downstream phase, and modifies theswitching matrix to activate the switching record for thenew connection. As soon as the acknowledge is receivedby the ingress node, the connection is established and thenew lightpath is modeled as a new direct edge in thevirtual topology multi-graph, henceforward called cutthrough edge (see [9]), with the capacity set to thedifference between the minimum capacity on all theedges belonging to the lightpath and the fraction of thelink bandwidth required by the involved connectionrequest. A cut-through edge can be used in any pathselection operation and thus can participate in one ormore groomed paths as a single virtual edge (a single hopat the IP layer). When an established lightpath is tomdown because the last connection occupying it is ended,the cut-through edge is removed and the edges in theextended graph corresponding to the underlying physicallinks are set back with full capacity. This schema allowsto flexibly model nearly any network topology and toconsider node conversion capabilities, wavelengthavailability and residual bandwidth per logical link at theIP layer. Each new connection request can be routed overa direct lightpath modeled as a single cut-through edge inour multi-graph, or over a sequence of lightpaths (amulti-hop path at the IP level, where each hop can be alightpath), if it crosses lambda-edge or routers as wellthat will link together lightpaths.

3.3. The traffic generation module

For each given WDM network topology, the trafficgenerator module randomizes source and destinationpairs according to a uniform distribution or a userprovided traffic matrix. Thus, traffic, expressed as thearrival of new connection request, may flow between anysource-destination pairs. Along its path, each connectionwill require some amount of the sub-wavelength unitsaccording to its bandwidth requirement. If therequirement of a call cannot be fulfilled when it arrives,the request will be blocked. The probability that a newrequest will be blocked is, thus, an important indicator ofthe overall efficiency provided by networks. Simulationis a rather indispensable method to estimate the blockingprobability, as well as other network performanceparameters. In large optical networks, call blockingprobabilities may be rare events due to the large capacityof the network or a very low request arrival rate. In suchcases, standard simulation may require an extremely longruntime, and usually incurs in large relative errors.Connection requests may arrive or be torn down at anymoment during the simulation. We developed a sessiontraffic source object which generates three types of

284


traffic, namely defined by a Poisson, exponential on/offand Pareto distribution with different average bit rate perwavelength. In particular, the Poisson distribution is avery good model for human requested end-to-endconnections (i.e. semi-permanent virtual circuits)whereas the Pareto one is very useful for modelingtypical Internet traffic-driven activity. In order to getmore confidence in the simulation results, we provide thepossibility to run multiple simulations on the samenetwork with the same algorithm and parameters andeventually get the average result values.

4. Implementation details

of design patterns. The simulator also includes anintuitive graphic user interface (GUI) allowing flexibledefinition and modification of network topology andsimulation parameters; a versatile configuration languageis used to define complex simulation environments. Theconfiguration file structure mainly consists of thephysical topology information, such as nodes and links,and the source/destination enabled nodes and theirpropertie s in terms of bandwidth range, its distributionand related parameters. After execution of the simulationa detailed trace file is generated which shows the exacttiming of the important events that have occurred duringthe course of simulation with the relevant details.

10ad (thousands of Connection requests)

Figure 1. Average rejection ratio on NSFNet.

5. Functional evaluation

9k 10k8k7k6k5k4k3k2kIk0.2

Ok

I I

I I

__ -'- __ -'- - - - - - - - - - - - - -'- - --I I I I

I I I 1 I I I

I . ' I I I I I I

- - -~ - -~ -:- - - ~ - - - ~ - - - ~ - - - ~ - - -~ - - -:- - --I T. ' I I I J I I I___ L __ J I J 1 ! L L I _

I I I I I I I t II I I I I J I I I

I I t I I I I I

- - - - - -1-- - -1- - - -1- - - -t - -- ~ - - - ~ - - -t- -- -1- - --

I I I I I I I I

I I I I I I I I______ ' 1 ..! .! 1. !.. ' 1 _

1 I I I 1 I I I

1 I I I I 1:+--7( MHA II I I I I I I I

0.4 - .... - - -1- - - -1- - - -< - - - • - - - .. - - - . :+--7( MlRA'- - - -I I I I I 1 I I

I I I I I I I )t----;t( :nvP I

OJ - -:- - - -:- - - -:- - - ~ - - - +- --T- - - ~ - - -:- - - -:- - --I t I I I I I I I

I I I t I I I 1 I_~ 1 1 ~ ~ ~ ~ ~ 1 _

0.9

0.8

0 0.7'.;:l."

" 0.600clJ 0.5u0

:0

We can immediately observe from the above Fig. 1the behavior, in terms of rejection ratio under increasingload, of some test RWA algorithms that closely matchesseveral results available in literature (e.g. [10)). We alsotested reliability of our simulator over variable networkspacing from ring to full-mesh topology, as in [5] (seeFig. 2). Starting from a simple 8 nodes ring network with8 fiber links and 32 wavelengths per link, new links areadded stepwise up to reach 32 links. At each step, eitherthe number of wavelengths on each link (series #1) or the

In order to validate the simulator, we ran manysimulations based on well-known algorithms on both testnetwork topologies and real networks such as NSFNetand Geant2, that have all been successfully modeled andtested. As performance measures, both throughput andblocking probability have been adopted. Figure I showsthe blocking probability for NSFNet network topologywith various known RWA algorithms.

I .---,.------,-----,----,---,---,.------,-----,----,------,

The simulator architecture has been realized inmodular form. This design choice stems from the natureof the problem and of the networks to be modeled. Eachmodule is implemented as a collection of cooperatingobjects and each network component is represented by aseparate object characterized by its own attributes andmethods. A small number of parameters can be easilydetermined from simulation for nondestructivemeasurements. Objects and modules have been used asthe building blocks of the whole simulation program. Allobjects have been designed in unified modeling language(UML), and implemented in Java. UML enabled to showthe block diagram of the simulator as well as thesimulation flow and the interaction among objects. Javahas been chosen in order to take advantage of its objectoriented paradigm, great extensibility, ease ofmodifiability, portability, and object serialization (tocheckpoint out simulations). Besides, Java provides alsothe garbage collector that automatically frees the unusedmemory, thus simplifying the code and optimizingmemory occupation. The simulator can run onworkstations with the Java Virtual Machine, e.g. MSwindows or various UNIX machines. The simulatorbuilds a model of the network in the multi-graph object,which is the main object representing the networktopology; it is made up of the sets of nodes, fibers, andedges (initially each edge corresponds to a singlewavelength). There is also the set of the paths currentlyrouted in the network that grows up as the connectionrequests are honored. Besides there is a set of supportobjects build up and maintained for efficiency sake: theseobjects comprise indexes, vectors and matrices that haveproven to speed up the computational time of thesimulations. Each object has been designed keeping inmind efficiency, usability and flexibility. Abstract objectsare used to uniform interfaces and are extended andinstantiated as concrete objects representing real networkelements. Object oriented encapsulation, inheritance andpolymorphism have been exploited thanks also to the use

978-1-4244-4671-1/09/$25.00 ©2009 IEEE 285


wavelength transport capacity (series #2) is decreased inorder to keep constant the network transport capacity,thus obtaining two series of test networks, as in Table 1.

for optical network simulations.

6. Conclusion

Figure 2. Ring to Full-Mesh test networks topologies.

Test series #1: ;. varying, OC-unit constantTest network 1 2 3 4Fiberlinks 8 16 24 32A. per link 32 16 10 8

OC-unit per A. 24/48 24/48 24/48 24/48

Test series #2: ;. constant, OC-unit varyingTest network 1 2 3 4Fiberlinks 8 16 24 32A. per link 4 4 4 4

OC-unit per A. 768 384 256 192

Simulators have proven to be an indispensable tool forthe study of the rapidly evolving optical networkstechnologies. In this paper we have presented SimulNet,a flexible simulator for WDM-routed networks. SimulNethas been expressly realized for the design and theevaluation of both RWA and optimization algorithms aswell as for the validation of signaling protocols andscheduling schemes. SimulNet has shown goodflexibility in managing even complex networks and hasexhibited accuracy of simulations results and satisfactoryperformances, making it an useful tool for the opticalnetwork research community.

7. References

Figure 3. Simulations on Ring to Full-Mesh topologies.

Table 1. Test networks topologies parameters.[I] Kim,Y., et aI., "GLASS (GMPLS Lightwave Agile

Switching Simulator) - A Scalable Discrete Event NetworkSimulator for GMPLS·based Optical Internet" , Proceedingsofthe lCCI '02 conference, 2002.

[2] Ogielski, A.T., SSFnet. Presentation at the DARPA NextGeneration Internet Conference , Arlington, VA, 1999.

[3] Wen, 8. , Bhide, N., Shenai , R., Sivalingam, K., "Opticalwavelength division multiplexing (WDM) networksimulator (OWns): Architecture and performance studies" ,SPIE Optical Networks Magazine, pp. 16-26,2001.

[4] The Network Simulator(ns), http://www.isLedu/nsnam /ns/.

[5] Pedrola, 0., Rumley , S., Klinkowski , M., Careglio, D.,Gaumier, C., Sole-Pareta, J., "Flexible Simulators for OBSNetwork Architectures", Proceedings of the IEEE ICTON2008 conference, 2008.

[6] Guerin, R., Williams, D., and Orda, A., "QoS RoutingMechanisms and OSPF Extensions", Proceedings ofGLOBECOM, 1997.

[7] Kar, K., Kodialam, M. and Lakshman, T. V., "Minimuminterference routing of bandwidth guaranteed tunnels withMPLS traffic engineering applications," IEEE J Sel. AreasCommun., vol. 18, no. 12, pp. 921-940, 2000.

[8] Kar, K. Kodialam, M. and Lakshman, T. V., "IntegratedDynamic IP and Wavelength Routing in IP over WDMNetworks", Proceedings of IEEE Infocom conference, 2001.

[9] Balasubramanian, S., Somani, A. K., "On Traffic GroomingChoices for IP over WDM networks," Proceedings of IEEEBroadnets conference, 2006.

[10] Boutaba, R., Szeto, W., Iraqi, Y., "DORA: EfficientRouting for MPLS Traffic Engineering", Journal ofNetwork and Systems Management, Vol. 10, No.3, 2002.

43

- - - - - - - - - - - - - - - - - - , .-

- - - - *-7< T.,t , . ri., =! (OC-unir-1-1)

*-7< T.'lS.ri., =! (OC-unir--lS)

>f--?< Test series =2

2Network topology

-----------

0.6"

0.5-1-------+-------t----------l1

Q7 ------------------------

;:§' 0.9 - - - ~

"8"".tt 0.8 - - - --"<J.Q,D

~g.~.:..8

~o

U

As expected, better performances are achieved with agood balance between the number of channels(wavelengths) and transport capacity (OC-unit) on eachchannel. The first network is penalized because of thefew links between node pairs, while the last because ofeither the small number of wavelengths per link (series#1) or the low transport capacity (series #2). Comparingthe results of tests over known real networks and theeffects of varying wavelengths and traffic load fordifferent network topologies, we can conclude that oursimulator is implemented properly and could be useful

The Dijkstra shortest path routing has been used toroute connection requests having a bandwidthrequirement between I and 12 OC-units and uniformdistribution between source/destination nodes isassumed. The results are shown in Fig. 3.

1.----------------------,

978-1-4244-4671-1/09/$25.00 ©2009 IEEE 286


simulnet: a wavelength-routed optical network simulation...

Documents