An

AS

a

ARA

KOOBBS

1

ettdiduta

iwwcstAttde

0d

Optik 122 (2011) 616–619

Contents lists available at ScienceDirect

Optik

journa l homepage: www.e lsev ier .de / i j leo

nalysis of burst/packet assembly techniques in high-speed optical switchingetwork

mit Kumar Gargchool of Electronics and Communication Engg, Shri Mata Vaishno Devi University (J&K), India

r t i c l e i n f o

rticle history:eceived 25 September 2009ccepted 30 March 2010

eywords:

a b s t r a c t

In the Optical Burst Switching (OBS) Network, the burst assembly technique is one of the challengingissues in the implementation of the system. It has the influence on the burst characteristic, which givesan impact on the network performance. Burst assembly is the process of assembling incoming datafrom the higher layer into bursts at the ingress edge node of the OBS network. The burst assemblymechanism must then place these packets into bursts based on some assembly policy. In this paper, the

ptical Burst SwitchingBS Networkurst assemblyurst droppingegmentation of burst

OBS system performance has been observed in simulated 12-node network based on Just-Enough-Time(JET) reservation protocol with various burst assembly techniques under the standard drop policy (DP)and the segmentation policy for contention resolution. The simulation results show that the performanceof the proposed Adaptive-Threshold with Fixed Maximum Time Limitation (ATH-FMTL) burst assemblyscheme is better than conventional burst assembly schemes in terms of loss probability and averageassembly delay. Also, the proposed scheme avoids a sudden increase in the burst size and makes the

s com

burst sent out smoother a

. Introduction

Photonic networks using Optical Burst Switching (OBS) hasmerged as an attractive choice for building the next genera-ion photonic Internet. In burst switched photonic networks, theransmission links carry multiple WDM channels, which can beynamically assigned to user data bursts. One channel on each link

s reserved for control information. This separation of control andata simplifies the data path implementation, facilitating greaterse of optical switching technologies. OBS combines the advan-ages of both circuit and packet switching (as shown in Table 1)nd ensures efficient bandwidth and resource utilization.

An OBS network consists of core nodes and end-devicesnterconnected by WDM fibers as shown in Fig. 1. In OBS net-

orks, the transmission links carry multiple WDM channels,hich can be dynamically assigned to user data bursts. One

hannel on each link is reserved for control information. Thiseparation of control and data simplifies the data path implemen-ation, facilitating greater use of optical switching technologies.t the edge of the network, a Setup control packet is sent on

he control channel to announce an upcoming burst. The con-rol packet is then followed by a burst of data after a shortelay. At the intermediate node, the Setup control packet islectronically processed, while the data channels are switched

E-mail address: garg amit03@yahoo.co.in.

030-4026/$ – see front matter © 2010 Elsevier GmbH. All rights reserved.oi:10.1016/j.ijleo.2010.03.025

pared to conventional schemes.© 2010 Elsevier GmbH. All rights reserved.

through transparently with no examination or interpretation of thedata.

Since OBS is based on statistical multiplexing, burst contentionsmay arise at the nodes, and thus resulting packet loss. Therefore,in order to make statistical multiplexing more efficient and also toreduce packet loss probability due to burst contentions, excessivevariation in burst size, in this paper, an efficient burst assemblyscheme named as Adaptive-Threshold with Fixed Maximum TimeLimitation Burst Assembly (ATH-FMTL) has been proposed, whichuses optimal burst length threshold and fixed maximum time lim-itation as the condition for burst generation. The burst lengththresholds are increased or decreased in case the burst queue size,at the time of burst generation, is larger than upper threshold orsmaller than lower threshold, respectively. The performance of pro-posed scheme is compared under the standard drop policy (DP) andthe segmentation policy (SDP) for packet loss probability.

The rest of the paper is organized as follows. In Section 2 a shortstate-of-the art in literature regarding conventional burst assem-bly methods such as timer-based and threshold-based have beendescribed. It is seen that by calculating the optimum thresholdvalue, calculating the minimum burst length and using a time-outvalue based on the packet’s delay tolerance, minimum packet losscan be obtained, while satisfying the delay requirement. The cur-

rent burst assembly mechanisms lack flexibility to actual networktraffic and increases assembly overhead and delay. Thus, in Section3 an efficient burst/packet assembly technique (ATH-FMTL) hasbeen proposed to achieve the equilibrium between the incoming

A.K. Garg / Optik 122 (2011) 616–619 617

Table 1Comparison of optical switching paradigms.

Optical switching (paradigm) Bandwidth utilization Latency (set-up) Optical buffer Traffic adaptively

HLL

pisSopsi

2

tneabtboaptmoait

2

aesbnietIrbtoea

Fd

Circuit LowPacket/cell HighBurst High

ackets to the transmission of the bursts. Achieving an equilibriummplies that the rate of arrival packets at the ingress node corre-ponds to the rate at which the bursts are formed and transmitted.ection 4 provides the brief methodology to verify the feasibilityf the proposed scheme. In Section 5 simulation results has beenresented along with the performance comparison of the proposedcheme (ATH-FMTL) with conventional schemes. Finally, conclud-ng remarks are made in Section 6.

. Burst assembly

Burst assembly is the process of assembling incoming data fromhe higher layer into bursts at the ingress edge node of the OBSetwork. As packets arrive from the higher layer, they are stored inlectronic buffers according to their destination and class. The burstssembly mechanism must then place these packets into burstsased on some assembly policy. The most common burst assemblyechniques are timer-based and threshold-based [1]. In timer-basedurst assembly approaches, a burst is created and sent into theptical network at periodic time intervals. In threshold-based burstssembly approaches, a limit is placed on the maximum number ofackets contained in each burst. Using both time-out and thresholdogether provides the best of both schemes and burst generation is

ore flexible than having only one of the above. By calculating theptimum threshold value, calculating the minimum burst lengthnd using a time-out value based on the packet’s delay tolerance,t is ensured that minimum loss can be obtained, while satisfyinghe delay requirement.

.1. Data burst generation

The OBS network structure components can be considered sep-rately as two groups; the edge nodes and the core nodes. Thedge node at the ingress side receives the IP packets from out-ide network and then assembles those IP packets into the burstsased on their destination and class of service. After that the edgeode will forward the control packet and data burst with the pol-

cy according to the reservation protocol. For the edge node at thegress side, it does the function of de-framing and de-assemblinghe data burst into multiple IP packets in a rather simple manner.t also handles the burst reordering and retransmission request ifequired. The core nodes have to perform the function of routing

urst scheduling, contention resolution and protection and restora-ion mechanism. Burst assembly process is done at the edge nodef the OBS network. It is the process that assembles several IP pack-ts into a burst grouping by their destinations and priorities. Thessembly buffer threshold and time threshold are two influences

ig. 1. An OBS network (electrical domain using aggregate traffic and the opticalomain using OBS).

igh Not required Lowow Required Highow Not required High

that make the effect to the burst generation. The assembly bufferthreshold is used to limit the maximum size of the burst, whilethe assembly time threshold is used to limit the maximum assem-bly waiting time of the burst generation. The conventional burstassembly schemes are described as follows.

2.1.1. Threshold-based without time limitation (TH)This threshold-based assembly scheme uses the fixed burst

size decision value. It does not include the maximum assemblydelay time limitation, thus it does not guarantee for the maximumassembly delay time. The burst will be transmitted when its lengthreaches the threshold value [2,3]. In case of constant packet size,the generated bursts will have the fixed size value. While packetsize varies, the generated bursts will have a varied burst size value.

2.1.2. Timers-based with minimum burst length (TM)The author in [4] proposed the timer-counter-based burst

assembly method. This method uses the concept of maximum wait-ing time and minimum burst length. The burst will be generatedas soon as the assembly waiting time reaches the maximum timelimitation. The size of the bursts must be equal to or bigger than agiven minimum burst size. If the waiting time is reached while theburst size is still smaller than the minimum burst size, the burstwill be padded to the minimum burst size and transmitted into thenetwork.

2.1.3. Fixed length threshold-based with time limitation (FL)A fixed length threshold-based assembly method uses both fixed

threshold and time-out interval for constant burst size and lim-ited delay time of the assembly function [5,6]. The timer starts tocount as soon as the first packet arrives at an assembly queue. Theburst consisting of IP packets in the assembly queue is sent outwhenever the time-out occurs or the number of packets in assem-bly queue reaches the threshold value. The additional technique,called padding, is used to make the generated burst size equal tothe designed fixed value.

3. Proposed assembly scheme (ATH-FMTL)

The current burst assembly mechanisms lack flexibility to actualnetwork traffic and increases assembly overhead and delay. Theexisting limit factors bring pressure to bear on core networkand make the network performance deteriorate. Assembly timeis inversely proportional to the offset time, that is, the longer theassembly time, the less time will be available for setting an off-set at the burst queue. The aim of the proposed scheme is toachieve equilibrium between the incoming packets to the trans-mission of the bursts. Achieving an equilibrium implies that therate of arrival packets at the ingress node corresponds to the rateat which the bursts are formed and transmitted. In order to alleviatean excessive variation in burst size, a novel burst assembly schemenamed as Adaptive-Threshold with Fixed Maximum Time Limita-tion Burst Assembly (ATH-FMTL) has been proposed, which uses

optimal burst length threshold and fixed maximum time limitationas the condition for burst generation. The burst length thresholdsare increased or decreased in case the burst queue size, at thetime of burst generation, is larger than upper threshold or smallerthan lower threshold, respectively. The performance of proposed

6 122 (

ssdwsitpatsB

1

2

3

4

5

6

7

tqd

4

i

high loss can be attributed to the loss of packets during the recon-figuration of a switch during contention resolution. The steepnessin the fall of packet loss is proportional to the switching time. Asthe switching time becomes insignificant with respect to the burstsize, the loss remains steady between the range 300 and 410 pack-

18 A.K. Garg / Optik

cheme is compared under the standard drop policy (DP) and theegmentation policy (SDP) for contention resolution. In the stan-ard drop policy, the later arriving burst is dropped if it contendsith another burst. In the segmentation policy, the overlapping

egments of earlier-arriving burst are dropped when the later arriv-ng burst contends with it [7]. Simulation results have shown thathe proposed scheme achieves a significant enhancement on theerformance of OBS networks in terms of burst loss rate with ancceptable delay. To reduce the burstiness of traffic by smoothinghe size of the bursts, the working of the proposed burst assemblycheme (Adaptive-Threshold with Fixed Maximum Time Limitationurst Assembly (ATH-FMTL)) is as follows:

. The packets arrive at the corresponding port and service classassembly queue becomes operative.

. To classify the packets into the appropriate burst, the decisionmaking is performed based on the fact that every packet has adelay tolerance that allows for flexibility during packet routingand on the assumption that no packet has a delay tolerance lessthat the amount of time it takes to route the packet through theOBS network, using the shortest route to it’s destination.

. Each burst length is estimated at the end of tp (prediction time)according to the past burst length value and current arrival traf-fic.

. Edge node determines the variable burst assembly duration(VBAD) by estimating burst size with current or previous load.

. Control packet is sent to OBS core network at time �;

� = ta − to(ta : assembly time; to : offset-time)

. It is assumed that wavelength conversion is available at everyLabel Switch Router (LSR) node in the core and a Just-Enough-Time reservation scheme is used.

. The Latest Available Unused Channel with Void Filling (LAUCVF)scheduling scheme is used.

The following pseudo code provides a more detailed descrip-ion of the proposed mechanism, where Qsd denotes the assemblyueue at ingress node s for destination d and VABDs is the assemblyuration used by node:

BeginPacket arrives at node s;d: destination of the new IP packet

If (Fixed assembly duration (Qsd) is not running) thenStart VABDs timer forQsd;

End ifAssemble the packet to corresponding queue Qsd;Update the burst length information;

If (Assembled burst Length ≥ Expected burst length) thenGenerate burst control packet for this burst;Fill in and send out the control packet on a control channel;Schedule the data burst to be sent out on a data channel

after an offset time;Stop the VABDs timer forQsd;The burst length margin is filled with void into the next

data burst;End if

End

. Methodology

To verify the correctness of the models, computer simulations a very useful and effective method. By comparing simulation

2011) 616–619

results with collected real data or mathematical analysis, corre-sponding parameters related to the performance of networks canbe modified. The network simulator NS-2 is a discrete event simu-lator targeted at networking research [8]. NS-2 provides substantialsupport for simulation of TCP, routing and multicast protocols overwired and wireless (local and satellite) networks. The details of thesimulation are as follows:

1. The network structure used in this study is the structure of theNSF Network with 12-nodes.

2. All the nodes have the function of both edge and core node,depending on which pair of node is selected to be the source(ingress edge node) and the destination (egress edge node).

3. Burst lengths up to 5000 bytes and a minimum 4000 bytes.4. There are no fiber delay lines in the network.5. Linux operating system.6. Simulation platform Ns-2.28.7. Programming languages C++ and TCL.8. Base simulator OBS.o.9a.9. The reservation scheme is based on the Just-Enough-Time (JET)

reservation protocol.10. The source and destination of each traffic flow are uniformly

selected among the nodes.11. Traffic is uniformly distributed over all source-destination

pairs.12. Average burst length of 100 �s.13. Transmission rate is 10 Gbps.14. Switching time is 10 �s.15. Fixed shortest path routing is used between all node pairs.16. Burst arrivals to the network are Poisson.17. The data bursts are not retransmitted.18. Bit errors in transmission are ignored.19. 95% confidence interval (batch mean method).

5. Simulation results

Fig. 2 plots the total packet loss probability versus the load forthreshold values of 100, 300 and 500 packets for both DP and SDP.It is observed that a threshold of 300 performs better than the othertwo selected threshold values, 100 and 500. Hence it is essential tofind an optimal threshold range to minimize loss. The need for opti-mal threshold can be better understood by analyzing Fig. 3. Here,it has been observed that the loss initially decreases, hits a mini-mum value and then begins to increase. The loss is minimal whenthe threshold value is between 300 and 410 packets. The initial

Fig. 2. Packet loss probability versus load.

A.K. Garg / Optik 122 (

Fig. 3. Packet loss probability versus threshold values.

ercpmt

oadsaflvad

Fig. 4. Loss probability versus offered load.

ts. After 410, the loss increases, since an increase in the thresholdesults in an increase in the average number of packets lost perontention. The optimal threshold value has been chosen as 300ackets for our network under a load range of 0–1 Erlang. The opti-al threshold may vary based on the nodal degree of the network,

he burst arrival rate and the load range of the network.Fig. 4 shows the plot of network loss probability against the

ffered load (0–1 Erlang) at 300 optimal threshold values runningt the bandwidth of 250 Mbit/s. Fig. 5 shows the average assemblyelay time of four assembly methods in the network level. It can beeen that the system performance in terms of loss probability andverage assembly delay time of each assembly method is clearly dif-

erent in the low offered load condition but behave equally in highoad circumstances. Even though the threshold-based method pro-ides the lowest loss probability value, it also produces the highestverage assembly delay in low load condition. This high assemblyelay time is obtained because there is no time limitation in the

Fig. 5. Average assembly delay versus offered load.

[

[

[

[

[

[

[

[

2011) 616–619 619

threshold-based technique. In the other three techniques, whichinclude the maximum delay time in their assembly function, theATH-FMTL scheme provides the lowest loss probability and assem-bly delay time among these three techniques. It can be concludedthat the adaptable threshold probability based assembly techniquereduces the system loss probability from the fixed-based technique.Even though the average loss probability values of all assemblyschemes are in the same range, there is a very high loss proba-bility variation on the time-based method, when compared to theothers.

6. Conclusion

In the proposed work, the assembly function of the edge switchin the Optical Burst Switching (OBS) Network with different assem-bly methods is analyzed. The main purpose of burst assembly is toincrease the switching efficiency at OBS core nodes. At the sametime, it has been shown that assembly algorithms can also smooththe input packet traffic and reduce the data loss to some extent. Itcan be seen that the system performance in terms of loss proba-bility and average assembly delay time of each assembly methodis clearly different in the low offered load condition but behaveequally in high load circumstances. From simulation results, it isclear that loss can be minimized by finding an optimal thresholdrange. The network loss probability performance of the threshold-based method provides the lowest loss probability but with thehighest maximum assembly delay time. The time-based, where theburst creation is based on the time threshold only, provides thehighest loss probability variation when compared to other threeassembly techniques and leads to wide variation of the qualityof service for individual connections. The fixed-based assemblyscheme provides the loss in the range between the threshold-basedand time-based techniques and the lowest average assembly delaytime among these three schemes. The proposed scheme improvesthe fixed-based method by using the adaptable burst size decisionvalue and reduces the system loss probability and average assem-bly delay. Furthermore, since the adaptable property depends onhow often the bursts are created by the time limitation criteria, thereduction in loss probability depends on the value of burst sizedecision threshold and the maximum assembly delay time. Theproposed assembly scheme assembles bursts more efficiently com-pared to conventional ones. It increases precision and reduces burstdelay. The proposed schemes vary burst sizes based on the linkcongestion levels while keeping end-to-end transmission delay ina feasible range. The scheme also ensures that sent bursts are nottoo short, which would compromise the performance of the OBSnetwork.

References

1] A. Ge, F. Callegati, L.S. Tamil, An optical burst switching and self-similar traffic,IEEE Commun. Lett. 4 (March (3)) (2000) 100–198.

2] X. Cao, J. Li, Y. Chen, C. Qiao, TCP/IP packets assembly over optical burst switch-ing network, in: Proceedings, IEEE Globecom, vol. 3, November, 2002, pp.2808–2812.

3] X. Yu, Y. Chen, C. Qiao, Study of traffic statistics of assembled burst traffic inoptical burst switched networks, in: Proceedings, SPIE Optical-Communication,2002, pp. 149–159.

4] X. Yu, Y. Chen, C. Qiao, Performance evaluation of optical burst switching withassembled burst traffic input, in: Proceedings, IEEE Globecom, vol. 3, November,2002, pp. 2318–2322.

5] Xu Lisong, Perros G. Harry, Rouskas George, Techniques for optical packetswitching and optical burst switching, IEEE Commun. Mag. 39 (January (1))(2001) 136–142.

6] Dolzer Klaus, Gauger Chirstoph, On burst assembly in optical burst switching

networks – a performance evaluation of just-enough-time, in: Proceeding ofthe 17th International Tele-traffic Congress, Salvador, Brazil, September 24–28,2001.

7] Oh Se-yoon, Hong HyunHa, Kang Minho, A data burst assembly algorithm inoptical burst switching networks, ETRI J. 24 (August (4)) (2002) 311–322.

8] Network Simulator, NS-2, available at http://www.isi.edu/nsnam/ns.