4-1 Experimental Study of a Burst-Switched WDM Network Testbed

79Yongmei Sun et al.

1 Introduction

The advances of wavelength division mul-tiplexing (WDM) technology and rapidgrowth of Internet traffic have generated aserious mismatch between huge transmissioncapacity of optical fibers and routing/forward-ing capability of electronic routers, which hastriggered many research activities on opticalswitching technologies. Among variousswitching technologies, optical burst switch-ing (OBS)［1］［2］ is considered an attractiveswitching paradigm because it is more effi-cient than optical circuit switching in terms of

wavelength utilization efficiency and has lessstringent requirements for optical devices thanoptical packet switching.

A data burst and its control signal, inburst-switched optical networks, are transmit-ted on separate channels and respectivelyswitched in optical and electronic domains.Burst is an aggregation of multiple client datawith the same egress node address; it is sentout following its control signal with a shortdelay called “offset time” without having towait for reservation acknowledgement (one-way reservation paradigm). The offset timeallows intermediate nodes to complete control

4 Wavelength Routing / Optical BurstSwitching / Optical Access Network

4-1 Experimental Study of a Burst-SwitchedWDM Network Testbed

Yongmei Sun, HASHIGUCHI Tomohiro, Vu Quang Minh, Xi Wang, IMAIZUMI Hideaki, MORIKAWA Hiroyuki, and AOYAMA Tomonori

Optical burst switching is considered an attractive switching technology for building the next-generation optical Internet. To investigate its feasibility, evaluate its performance and explore itsfuture direction, we designed and implemented an overlay-mode optical burst-switched networktestbed. In this report, we present the node architecture, control algorithm, and performanceevaluation of the testbed. A flexible “ transceiver + forwarding” node architecture is proposed toperform both electronic burst assembly/disassembly and optical burst forwarding. It is designedto provide class of service and wavelength selection for locally generated bursts, and trans-parency to cut-through bursts. A scheduling mechanism, which efficiently combines two differentcontention resolutions in space and wavelength domains, is discussed in detail. Performancesof the burst-switched network testbed, including end-to-end delay, burst blocking probability andTCP throughput, are evaluated; and online video services are demonstrated. Furthermore, keydeterminants of the network performance and future directions are also discussed.

KeywordsOptical burst switching, Burst assembly, Wavelength selection, Burst scheduling,Contention resolution

80 Journal of the National Institute of Information and Communications Technology Vol.53 No.2 2006

signal processing and optical switch configu-ration ahead of the burst arriving; thus no opti-cal buffers are necessary at intermediatenodes.

Since many key issues with OBS, such asburst assembly, signaling, scheduling and con-tention resolution, have been extensively stud-ied［3］-［8］, it is necessary and important todevelop a network testbed for evaluating OBStechnology. Up to now, several burst-switchednode prototypes and testbeds have been devel-oped and demonstrated［9］-［14］. However,most of them mainly focused on some keyunit technologies, such as optical switches andsignaling protocols, and can not provide a net-work-wide experimental platform with com-prehensive OBS functions. To investigate thefeasibility of OBS, evaluate OBS protocolsand algorithms, and study the future direction,we have designed and implemented a general-purpose and flexible OBS network testbed. Inthis report, node architecture, control algo-rithm, and performance evaluation of the test-bed are discussed. A flexible “transceiver +forwarding” node architecture is presentedfirst, which can perform transparent opticalburst switching and electronic burst assemblywith support for class of service (CoS) andwavelength selection. Then we discuss thedesign of a scheduling scheme mechanism,which efficiently combines two different con-tention resolutions in space and wavelengthdomain. Through a series of experiments, per-formances of the OBS network testbed areevaluated and discussed finally.

2 Design principles of the OBSnetwork testbed

An ideal testbed is expected to emulatereal OBS networks to the maximum, provide aflexible network-wide experimental platformfor new ideas and novel technologies, andcontribute a viable evolution solution to thenext-generation optical Internet. The follow-ing key functional requirements especiallyneed to be fulfilled: burst assembly and disas-sembly, network protocols (e.g., routing and

signaling), control algorithms (e.g., schedulingand contention resolution), scale and resources(e.g., sufficient nodes, WDM links, and wave-lengths), and compatibility with legacy net-works (e.g., interconnection with the Internet).To meet such challenging requirements, a gen-eral-purpose and flexible OBS testbed wasdesigned. Detailed principles that we consid-ered in the design are as follows:

(1) Compatibility and interoperability with IPnetworksConsidering the popularity of IP networks,

a testbed ideally needs to be designed to per-form asynchronous variable-sized burstswitching to match the natural characteristicsof IP traffic. Furthermore, an overlay-modeneeds to be adopted. This means the testbedcan provide OBS functions for IP networkswithout any changes to them, promising goodcompatibility and interoperability with exist-ing IP networks and the benefits of networkevolution to the next-generation optical Inter-net.

(2) Generality, modularity, and expandabilityTo provide an open evaluation environ-

ment, a testbed needs to be designed so that itis general-purpose, which is achieved with atransparent data plane and a reprogrammablecontrol plane. It can support various traffic,protocols, and algorithms, e.g. flow/label/wavelength switching for IP or other traffic,just-enough-time (JET)［2］and just-in-time(JIT)［15］protocols. An OBS node needs to bedivided into multiple functional modules.Each of these, such as the switch matrix,needs to be integrated into an individual cir-cuit board. Higher performance devices, e.g.faster optical switches, can be employed bysimply replacing the corresponding board. Inaddition, a reconfigurable topology and suffi-cient resources are also important factors inenhancing the generality of the testbed.

Guided by the principles above, we imple-mented a general-purpose OBS networktestbed［16］with flexible node architecture,efficient JET signaling protocol, and novel


scheduling mechanism. The testbed consistsof one core node and three edge nodes. The IPnetwork is connected to the OBS networkthrough edge nodes. At the ingress edge node,multiple IP packets with the same egress edgenode address are assembled into one burst.Then, the burst is routed and forwardedthrough the OBS network. Finally, it is disas-sembled back into IP packets at the egressedge node. OBS nodes transmit and switchbursts in the data plane, and exchange andprocess control signals in the control plane.Since the core node is a simplification of theedge node, we will mainly discuss the edgenode hereafter.

3 “Transceiver + forwarding” nodearchitecture

In many previous works, edge nodes onlyperform burst assembly/disassembly at theboundary of OBS networks. To support vari-ous network topologies and more intelligence,however, we propose a flexible “transceiver +forwarding” architecture for edge nodes toforward cut-through bursts, as well as performburst assembly/disassembly with support forCoS and wavelength selection. As depicted inFig. 1, an edge node consists of an opticalswitching unit, a burst transceiver unit, and acontrol unit. The optical switching unit per-forms all-optical burst switching. It is com-posed of couplers, mux/demuxes, an opticalswitch matrix, power equalizers, and ampli-fiers. The burst transceiver unit performs burstassembly/disassembly. The control unitprocesses control signals, configures the opti-cal switch matrix, and controls the burst trans-ceiver unit.

An edge node supports two remote WDMlinks and four local links as shown in Fig. 1.For simplicity, we use the term remote link todenote a WDM link between two OBS nodes,which carries four DWDM burst channels(wavelengths) and one shared control channel(wavelength). Similarly, the term local linkdenotes a link between an OBS node and itsclient IP network, which carries only one

packet channel without employing WDMtechnology. For remote links, the control sig-nals are extracted from bursts by couplers andprocessed in the control unit, while de-multi-plexed bursts are sent to the switch matrix.After equalization, multiplexing, and amplifi-cation, bursts and control signals are multi-plexed into links by couplers. For local links,multiple Ethernet frames are first assembledinto bursts at the burst transmitter. Then theoptical bursts are sent to the switch matrix.These two types of bursts, referred to asremote and local, respectively, cut through theswitch matrix to their next hop nodes or localburst receiver according to the routing infor-mation contained in their control signals.Upon receiving a burst through the localswitch matrix, the burst receiver disassemblesit into multiple packets, which are forwardedto the legacy IP networks in a conventionalway.

3.1 Key issues in the optical switchingunit

The key point of designing the opticalswitching unit is how to implement a high-performance transparent optical path withconsiderations of optical device limitationsand physical-layer constraints. This goal isachieved by commercial high-performanceoptical devices incorporating carefullydesigned control circuits. Among them, opti-cal switch and power imbalance are two cru-cial concerns.

Fig.1 Node architecture.


There are various optical switches applica-ble nowadays［17］. However, consideringspeed, scale, and reliability, commercial PLCswitches were adopted to construct the 16×16non-blocking switch matrix. The switchingspeed is less than 3 ms and the insertion loss isless than 8 dB.

In OBS networks, power imbalance occursnot only between different links and wave-lengths but also within each wavelengthbecause the bursts within a wavelength maybecome from different sources and go throughdifferent nodes and paths. Such rapid powerfluctuations lead to instability and failure inserious cases. To solve this problem, wedesigned a dynamic channel-level powerequalization scheme. By using a magneto-optical variable optical attenuator (VOA)array and a carefully designed feedforwardcontrol circuit, power equalization could becompleted within 3 ms, which includes theresponse time of the VOA and feedforwardprocessing time.

3.2 Burst transceiver supporting CoSand wavelength selection

Considering the importance of CoS andthe wavelength assignment issue in OBS net-works, we focused our attention on how toefficiently and flexibly support CoS andwavelength selection in the design of the bursttransceiver unit. As shown in Fig. 2, a new“3-level FIFOs + 2-level switch” burst trans-mitter architecture is proposed and implement-ed with an Altera high-end FPGA. The three-level first-in first-out buffers (FIFOs) are usedfor routing, burst assembly, and scheduling.The first-level switch classifies burst queuesby destination and CoS, while the second-level switch executes wavelength selection.All the wavelengths can be shared for anyburst, which makes the edge node flexible insupporting various wavelength assignmentalgorithms, e.g. random and priority-basedassignment［18］. To simplify implementation, aburst is designed to be a simple aggregation ofmultiple Gigabit Ethernet frames with thesame egress node address, which could further

be classified by CoS. Asynchronous variable-sized bursts are generated to match the naturalcharacteristics of IP traffic. Furthermore,asynchrony simplifies implementation byeliminating synchronization and burst align-ment. More precisely, the burst transmitterworks as follows:(1) Gigabit Ethernet frames are buffered in the

first-level FIFOs and then switched to thesecond-level FIFOs according to theiregress node addresses and CoS attributes.

(2) Multiple frames buffered in the same sec-ond-level FIFO are assembled into oneburst according to a time-size-basedassembly mechanism［3］. Precisely, a burstis generated when either the assembly timethreshold or the burst size threshold isreached.

(3) After channel scheduling, the output wave-length channel and sending time are deter-mined. The generated burst is buffered inthe third-level FIFOs and transmitted at thescheduled time.

4 Contention resolution and burstscheduling

Since burst transport is in the one-wayconnectionless manner and optical RAM isnot yet available, contention resolution andburst scheduling are more challenging in OBSnetworks. Although fiber delay line and wave-length conversion have been proposed as con-

Fig.2 Burst transmitter and control unit.


tention resolutions, the immaturity of the tech-nologies and the absence of a full configura-tion for OBS networks have necessitated thestudy of other approaches. A simple deflectionrouting protocol［19］and a smart wavelengthassignment algorithm, called priority-basedwavelength assignment (PWA)［18］have beenproposed to efficiently decrease contentionsand reduce burst blocking probability. Asshown in Fig. 3, the deflection routingapproach provides a detour path for a contend-ing burst. With PWA, each edge node main-tains a dynamically updated wavelength prior-ity database where every wavelength is priori-tized for each destination by learning from itsutilization history. More specifically, when anode receives an ACK, which indicates a suc-cessful burst delivery, it increases the priorityof the corresponding wavelength. Otherwise,on receiving a NACK, which indicates a burstloss, it decreases the priority of the corre-sponding wavelength. By prioritizing wave-lengths, nodes tend to assign different wave-lengths to bursts sharing the same links of thenetwork, therefore proactively avoiding colli-sions as much as possible.

The two approaches above are efficientlycombined in a carefully designed burst sched-uling algorithm in our OBS testbed. Using theburst status information carried in control sig-nal to indicate the deflection status and failurereason for each burst, wavelength prioritycould be properly updated while exploitingdeflection routing functionality. By compre-hensively managing the wavelength priority

database, scheduling information tables, andforwarding procedure, the burst blockingprobability is decreased and bandwidth utiliza-tion is improved. Details on the burst schedul-ing procedure are described below.

Scheduling information is saved in multi-ple scheduling tables. For an edge node withM output links and every link carrying Nchannels, there are M×N scheduling tablesdenoted CST(m , n), where 1≤m≤M and1≤n≤N. Each CST(m, n) contains schedulinginformation in a period of time, TT, includingthe start and end times of each scheduledburst. A scheduler manages CST(m, n) andkeeps track of the unscheduled time.

Figure 4 shows the burst scheduling proce-dure. For a local burst, the procedure can bedivided into the following steps: (1) When a burst is generated, its output link L

is obtained. The scheduler looks up CST(L,n) (1≤n≤N) and finds an available outgoingchannel, which is operated in order ofchannel priority. If one such channel, C,exists, go to Step 3; otherwise go to Step 2.

(2) The scheduler looks for an available out-going channel in the near future within TT.If it succeeds, go to Step 3; otherwise theburst is discarded.

(3) Output channel C is assigned and schedul-ing table CST(L, C) is updated.

(4) The burst is buffered and sent at the sched-uled time.For a remote burst, its output link L, and

channel C are obtained by interpreting its con-trol signal. Thus, the scheduler looks up

Fig.3 Concepts of deflection routing andPWA. Fig.4 Scheduling procedure.


CST(L, C) and tries to find whether it is avail-able. If successful, CST(L, C) is updated andthe burst is forwarded or received; otherwisethe burst is deflected or discarded.

This scheduling mechanism needs O(EN)and O(MN) memories to record the channelpriority and scheduling information, respec-tively, where E is the number of edge nodes. Italso takes O(logN) time to maintain a rankedpriority list, and O(N) time to search a suitablewavelength. The required memory capacity issensitive to the scale of the network. However,a large-capacity memory is no longer expen-sive and is easily available. The time for main-taining the priority list and searching thewavelength only depends on the number ofwavelengths regardless of the number ofnodes.

5 Evaluation and discussion

We built an optical burst-switched networktestbed with self-developed OBS nodes; itsmain specifications are listed in Table 1.

On this OBS network testbed, importantperformance metrics, including end-to-enddelay, burst blocking probability and TCPthroughput, were evaluated and discussed.

Furthermore, online video services were alsodemonstrated. Figure 5 outlines the experi-mental configuration of the OBS network.Three edge nodes were connected with 20 kmfibers to form a ring network, where burstswere transferred in a clockwise direction. Inexperiments of delay and blocking probability,three clients were emulated with an AgilentRouter Tester.

5.1 End-to-end delayWe define end-to-end delay, D, as the

average time taken by IP packets from enter-ing an OBS network to exiting it. More specif-ically, it is composed of four parts:

Din: delay introduced at the ingress node; TO: offset time between a burst and its

control signal; Dt: propagation delay; De: delay introduced at the egress node.

As a distinct feature of OBS, offset time isan important factor of the end-to-end delay.Therefore we discuss it first. Considering thenode architecture and characteristics of opticaldevices aforementioned, we introduce a guardtime in one-way signaling procedure to com-pensate for the response time of the opticalswitch and power equalizer. In other words,bandwidth reservation starts ahead of burstarrival with the guard time. In our OBS test-bed, the measured guard time was 10 ms. Toleave a margin for control signal processingand to make the transmission more reliable,

Table 1 Specifications of the OBS testbed

Fig.5 Configuration of the experimentalnetwork.


the offset time TO was set to 13 ms in ourexperiments.

Burst assembly has a great effect on theend-to-end delay. Hence, we measured theend-to-end delay between clients A and C as afunction of burst assembly time, which wasplotted in Fig. 6. Due to Dt =0.2 ms and De <1 ms in this experiment, the delay introducedby the ingress node was also approximatelyplotted in the same figure. We noted that Din

was shorter than the burst assembly time. Thisphenomenon is due to the fact that only thefirst packet must wait for the duration time ofthe whole assembly procedure for each burst,while the other packets wait for a shorter time.We also observed that the end-to-end delay ismainly contributed from the ingress node andguard time. The former mainly results fromburst assembly and scheduling while the lattermainly results from the performance of theoptical switch matrix.

5.2 Burst blocking probabilityBurst blocking probability is usually used

to evaluate the effectiveness of contention res-olution. In this experiment, burst blockingprobability was measured in three cases: nocontention resolution, PWA only, and deflec-tion only. Each client sent IP packets to anoth-er two clients. Generated bursts were trans-ferred in a clockwise direction. While in thecase of deflection routing, contending burstswere deflected in a counterclockwise direc-tion. Due to the short distance between nodes,the second collision between two burstsoccurred at the destination node is not consid-ered.

Experimental results are plotted in Fig. 7,where the network traffic was measured interms of bursts generated by all nodes and wasexpressed as an average number of bits persecond. By adopting PWA and deflection rout-ing separately, the burst blocking probabilitywas reduced from that of no contention resolu-tion, especially when the network traffic is nothigh. We note that deflection routing had amore remarkable effect than PWA. This isbecause the deflection routes have lower traf-fic load than the default routes. Therefore,most deflected bursts can be delivered to theirdestinations.

5.3 TCP throughputIt is important to study the performance of

TCP in OBS networks since TCP traffic is andmay remain to be the most popular traffic typein the future Internet. TCP provides a reliabletransport layer over an unreliable networklayer mainly by using slow start, congestionavoidance and retransmission mechanisms.In OBS networks, these mechanisms will beaffected by burst assembly algorithm, burstloss rate, and burst loss pattern as describedin［20］［21］. In brief, higher delay, loss rate andretransmission introduced by OBS networkswill degrade TCP throughput. On the otherhand, the fact that one burst contains multiplepackets enables TCP to reach a larger sendingwindow, and accordingly increases TCPthroughput.

We experimentally studied the relationshipbetween TCP throughput and burst assembly

Fig.6 Delay vs. burst assembly time.

Fig.7 Burst blocking probability vs. net-work traffic.


time. Figure 8 shows the theoretical through-put and available TCP throughput for a singlewavelength between client A and C. It indi-cates the degradation of TCP performanceover an OBS network. Another observation isthe existence of an optimal assembly time.This phenomenon can be explained as follows.A shorter assembly time leads to more burstloss within edge node due to the long guardtime, while a longer assembly time introduceslarger delay. Both of two cases decrease TCPthroughput compared with the case of optimalassembly time.

5.4 Application demonstrationAs shown in Fig. 9, online real-time trans-

mission of video stream data was demonstrat-ed on the OBS testbed. In this demonstrationthree clients were three personal computersand node 2 was connected with the Internet.Two real-time video stream services were suc-cessfully demonstrated simultaneously. Thefirst was a TCP-based video on demand ser-

vice between A and B. The second was aUDP-based live video chat service between Aand C using Windows Messenger through theInternet. Dynamical switching of these twovideo streams could be monitored at node-2.Interoperability with IP networks and the pos-sibility of providing TCP/UDP-based latency-sensitive video services over the OBS testbedwere verified.

5.5 DiscussionExperimental results above reveal that the

burst assembly, optical switching, and con-tention resolution are key determinants ofOBS network performance. End-to-end delayincreases slowly with the number of interme-diate nodes because of one-way signaling andcut-through burst switching. This greatly ben-efits scalability. In addition, faster opticalswitches and optimized burst processing pro-cedures are expected to effectively shorten thedelay. These two approaches can also improvebandwidth efficiency, another significant per-formance metric. Bandwidth efficiency isdefined as the ratio of burst size (in timedomain) to the sum of burst size and guardtime. In the current implementation, burst size,which is limited by the memory capacityavailable within the FPGA, is small comparedwith the long guard time. This leads to lowbandwidth efficiency. However, by employinga specific high capacity memory and adoptinghigh speed switches, bandwidth efficiency canbe dramatically improved. Due to the modulardesign of OBS nodes, the discussed improve-ments in performance can be implemented byreplacing the burst transceiver and switchboards. Furthermore, to achieve lower block-ing probability, sparse wavelength converters,optical buffers and other new ideas should beconsidered.

6 Conclusion

In this report, the design, implementation,and experiments of a general-purpose OBStestbed were presented. This testbed supportsvarious types of traffic and protocols, and

Fig.8 Throughput vs. burst assembly time.

Fig.9 Application demonstration on theOBS testbed.


enables simple system update with advanceddevices. In addition, important design parame-ters (e.g., assembly time threshold) can easilybe adjusted for various experimental studies.On this testbed, performances were evaluatedand discussed, and online video services weredemonstrated. To the best of our knowledge,this is the first implementation and demonstra-tion of a network-wide testbed with compre-hensive OBS functions. By improving its per-formance further, we can expect to provide aviable solution to the future optical Internet.

* Part of this paper has been published inIEEE Commun. Mag., Vol.43, No.11,Nov. 2005.

Acknowledgments

This work was supported by the NationalInstitute of Information CommunicationsTechnology (NICT). The authors would like tothank NTT Electronics Corporation for theirvaluable discussions and technical assistance.

References01 J. S. Turner, “Terabit Burst Switching”, J. High Speed Networks, Vol. 8, No. 1, pp. 3-16, Jan.1999.

02 C. Qiao and M. Yoo, “Optical Burst Switching (OBS) - A New Paradigm for an Optical Internet”, J. High

Speed Networks, Vol. 8, No. 1, pp. 69-84, Jan. 1999.

03 Y. Xiong, M. Vandenhoute, and H. Cankaya, “Control Architecture in Optical Burst-Switched WDM Net-

works”, IEEE JSAC, Vol. 18, No. 10, pp. 1838-1851, Oct. 2000.

04 L. Xu, H. G. Perros, and G. N. Rouskas, “Techniques for Optical Packet Switching and Optical Burst

Switching”, IEEE Commun. Mag., Vol. 39, No. 1, pp. 136-142, Jan. 2001.

05 V. M. Vokkarane, J. P. Jue, and S. Sitaraman, “Burst Segmentation: An Approach for Reducing Packet

Loss in Optical Burst Switched Networks”, Proc. ICC 2002, New York, NY, Apr. 2002.

06 S. Ovadia, C. Maciocco, M. Paniccia, and R. Rajaduray, “Photonics Burst Switching (PBS) Architecture

for Hop and Span–Constrained Optical Networks”, IEEE Commun. Mag., Vol. 41, No.11, pp. s24-s32,

Nov. 2003.

07 J. Li, C. Qiao, and Y. Chen, “Recent Progress in the Scheduling Algorithms in Optical-Burst-Switched

Networks”, J. Optical Networking, Vol. 3, No.4, pp. 229-241, Apr. 2004.

08 Y. Chen, C. Qiao and X. Yu, “Optical Burst Switching: A New Area in Optical Networking Research”,

IEEE Network, Vol. 18, No.3, pp. 16-23, May/Jun. 2004.

09 F. Masetti et al., “Design and Implementation of a Multi-Terabit Optical Burst/Packet Router Prototype”,

Proc. OFC 2002, pp. FD1-1-FD1-3, Anaheim, CA, Mar. 2002.

10 I. Baldine, M. Cassada, A. Bragg, G. Karmous-Edwards, and D. Stevenson, “Just-in-Time Optical Burst

Switching Implementation in the ATDnet All-Optical Networking Testbed”, Proc. Globecom 2003, pp.

2777-2781, San Francisco, CA, Dec. 2003.

11 A. Sahara et al., “Demonstration of Optical Burst Switching Network Utilizing PLC and MEMS Switches

with GMPLS Control”, Proc. ECOC 2004, pp. 896-897, Stockholm, Sweden, Sep. 2004.

12 K. Kitayama, M. Koga, H. Morikawa, S. Hara, and M. Kawai, “Optical Burst Switching Network Testbed

in Japan”, Proc. OFC2005, OWC3, Anaheim, USA, Mar. 2005.

13 H. Guo et al., “A Testbed for Optical Burst Switching Network”, OFC2005, OFA6, Anaheim, USA, Mar. 2005.

14 H. Yoo et al., “Microsecond Optical Burst Add-Drop Multiplexing for WDM Ring Networks”, Proc.

COIN2005, pp. 78-80, Chongqing, China, May-Jun. 2005.

15 I. Baldine, G. Rouskas, H. Perros, and D. Stevenson, “JumpStart: a Just-in-Time Signaling Architecture

for WDM Burst-Switched Networks”, IEEE Commun. Mag., Vol.40, No.2, pp. 82-89, Feb. 2002.


16 Y. Sun, T. Hashiguchi, V. Minh, X. Wang, H. Morikawa, and T. Aoyama: “Design and Implementation of an

Optical Burst-Switched Network Testbed”, IEEE Commun. Mag., Vol. 43, No. 11, pp. s48-s55, Nov. 2005.

17 X. Ma and G. S. Kuo, “Optical Switching Technology Comparison: Optical MEMS vs. Other Technolo-

gies”, IEEE Commun. Mag., Vol.41, No.11, pp. s16- s23, Nov. 2003.

18 X. Wang, H. Morikawa, and T. Aoyama, “Priority-based Wavelength Assignment Algorithm for Burst

Switched WDM Optical Networks”, IEICE Trans. Commun., Vol.E86-B, No.5, pp. 1508-1514, May 2003.

19 X. Wang, H. Morikawa, and T. Aoyama, “Burst Optical Deflection Routing Protocol for Wavelength Rout-

ing WDM Networks”, Opt. Net. Mag., Vol. 3, No. 6, pp. 12-19, Nov./ Dec. 2002.

20 A. Detti and M. Listanti, “Impact of Segments Aggregation on TCP Reno Flows in Optical Burst Switch-

ing Networks”, Proc INFOCOM, Vol. 3, pp. 1803–1812, 2002.

21 X. Yu, J. Li, X. Cao, Y. Chen, and C. Qiao, “Traffic Statistics and Performance Evaluation in Optical Burst

Switched Networks”, J. Lightwave Technology, Vol. 22, No. 12, pp. 2722-2738, Dec. 2004.

Yongmei Sun, Dr. Eng.

Department of Information and Com-munication Engineering, The Universi-ty of Tokyo

Photonic Network

Vu Quang Minh

Department of Information and Com-munication Engineering, The Universi-ty of Tokyo

Photonic Network

HASHIGUCHI Tomohiro, Dr. Sci.

Department of Frontier Informatics,The University of Tokyo

Photonic Network

Xi Wang, Dr. Eng.

Research Associate, Department ofInformation and Communication Engi-neering, The University of Tokyo

Photonic Network

IMAIZUMI Hideaki, Ph.D

Research Associate, Department ofInformation and Communication Engi-neering, The University of Tokyo

Internet Architecture, Routing, Photon-ic Network

MORIKAWA Hiroyuki, Dr. Eng.

Associate Professor, Department ofFrontier Informatics, The University ofTokyo

Ubiquitous Network, Mobile Network

AOYAMA Tomonori, Dr. Eng.

Professor, Department of Informationand Communication Engineering, TheUniversity of Tokyo

Photonic Network, Ubiquitous Network

Documents

4-1 Experimental Study of a Burst-Switched WDM Network Testbed