Study of Qos Performance in Optical Burst Switched Networks (OBS)

Transcription

1 Indian Journal of Science and Technology, Vol 9(48), DOI: /ijst/2016/v9i48/99269, December 2016 ISSN (Print) : ISSN (Online) : Study of Qos Performance in Optical Burst Switched Networks (OBS) Suriani Mohd Sam*, Salwani Mohd Daud, Kamilia Kamardin and Nurazean Maarop Advanced Informatics School, Universiti Teknologi Malaysia, Kuala Lumpur, Malaysia; suriani.kl@utm.my, salwani.kl@utm.my, kamilia@utm.my, nurazean.kl@utm.my Abstract One of the main challengers of the future Internet is to ensure QoS for various types of traffic characteristics with differentiated service requirements. It has been highlighted that OBS can provide high bandwidth data transmission in alloptical Internet network. However, the provisioning of QoS in OBS network is still an issue. The focused of this study is to develop QoS provisioning mechanism at the ingress and core node of OBS network. The proposed QoS provisioning ensures that the differentiated traffic is able to achieve their respective demands. This study proposed a DiffServ-aware ingress node with priority queueing (PQ) and burst assembly mechanisms. In addition, a core node with contention resolution scheme using FDL_WA_WPremp is developed as part of the integration of the proposed OBS network. Keywords: Burst Loss, Contention Resolution, FDL, OBS, Preemption, Quality-of-Service (QoS), Wavelength Assignment 1. Introduction Optical burst switching was introduced for optical WDM networks. Circuit switching employs two-way reservation method that has a large round trip. Packet switching alternatively, has a large buffer requirement, complex control and strict synchronization issues. OBS is designed to achieve a balance between the coarse-grained circuit switching and fine-grained packet switching. As such, a burst may be considered as having an intermediate granularity as compared to circuit and packet switching. Packets delivered at an OBS ingress node that are meant for the same egress OBS node and belong to the similar QoS class are combined and sent in bursts. At intermediate nodes, the data within the optical signal is transparently switched to the subsequent node according to forwarding information contained inside a control packet preceding the burst. At the egress node, the burst is subsequently de-assembled and sent out electronically. Given that optical burst switched networks is using connectionless transport, there exists the likelihood that bursts might contend with one another at intermediate nodes. Contention will happen if multiple bursts from different input ports are intended for the same output port at the same time. Contention resolution schemes may be used to provide QoS in an all-optical core network. Since OBS networks use connectionless transport, the bursts will have the possibility to contend with one another at the intermediate nodes. Contention happens when more than one burst attempt to reserve the same wavelength channel on an outgoing network. Burst loss that is caused from contention is a key issue in OBS networks. Such contention losses which are temporary in nature can lower the performance at the upper layers. In electronic network, contention is resolved by buffering the contending bursts. In spite of this, optical buffers are complicated to apply and there is no optical equivalent of random access memory. *Author for correspondence

2 Study of Qos Performance in Optical Burst Switched Networks (OBS) In OBS network, when contention occur one of the contending data burst is allowed to reserve the channel, for other data bursts one or a combination of the contention resolution techniques can be applied. If contention still unable to resolved, therefore one of the contending data burst is dropped. There are four main ways of resolving contention in OBS. The first two are the use of fiber delay lines (FDL) to delay the burst for a discrete and finite period 1-3 and deflecting the burst to another output fiber, known as deflection routing 4-7. The third resolution is by implementing wavelength conversion 8,9 where a contending burst is sent on another wavelength through wavelength conversion. The forth is the burst segmentation which a contending burst is broken into segments and the overlapped part of the contending burst is dropped. Traffic in a network is consists of flows spring through a diversity of applications on end stations. Their services and performance requirements are differed by their applications. Any flow s requirements rely on fundamentally on the application it is associate with. The network s capability to provide service required by specific network applications with certain degree of control over performance measures is classified into three service levels. The first category is the BE service. This is a fundamental connectivity without guarantee as to whether or when a packet is transmitted to the destination. When the input or output buffer queues are used up or when a contention happens in the OBS network, a packet is usually been discard. BE service is not literally a part of QoS since there is no service or delivery guarantees are accomplished during forwarding the best-effort traffic. The second category of QoS level is the DiffServ. Traffic is grouped into classes depending on their service requirements in DiffServ. Each traffic class is differentiated by the network and serviced accordingly to configure QoS mechanisms for a particular class. This QoS technique performs well for bandwidth intensive data applications. In order to ensure basic network connectivity at all time, prioritizing and differentiating the network control traffic from the rest of the data traffic is important. The final type of QoS level is the guaranteed service. This is a service that requires network resource reservation to ensure that the network meets a traffic flow s specific service requirements. Multimedia applications such as audio and video are the applications that require such services. QoS implementation is aimed to provide a connection with certain performance limits from the network. The performance of each QoS will be compared based on the two different burst assembly techniques, i.e. fixed timer burst and hybrid burst. A packet level study is conducted based on the traffic that enters into the network. The study is focused on looking at bandwidth utilization, average end-to-end packet delay and packet throughput. The simulation work involves the EF premium; AF assured and BE best effort traffic. The paper is organized as follows. In Section 2, the proposed QoS enhancement in OBS is briefly discussed. In Section 3, the analytical approach used in the simulation is presented. The propose network model and the simulation environment is presented in Section 4 meanwhile Section 5 discusses on the simulation results and its performance. The conclusion of this paper is presented in Section Proposed QoS Enhancement in OBS The main technology restrictions on optical edge and core switching are the lack of optical memories and the limitations of electronic/optical conversion devices. Therefore, in this work, Fiber Delay Line (FDL) is proposed to delay the burst at the edge nodes when no wavelengths are available during the limited time period. To further reduce the burst loss probability, the FDL technique is integrated with wavelength assignment and preemption schemes that are being deployed at the core nodes to achieve better performance. The propose scheme is known as FDL_WA_WPremp, which the performance will be analyzed simultaneously into the network model proposed Fiber Delay Line (FDL) at the Ingress Router In connectionless transport environment, competition among the burst in the OBS networks occurs at the intermediate nodes. Contention happens if more than one burst attempt to reserve the same wavelength channel on an outgoing link. In other words, burst loss due to contention is a main issue of interest in OBS networks. Although these losses are momentary, but it can degrade the performance at the higher 2 Indian Journal of Science and Technology

3 Suriani Mohd Sam, Salwani Mohd Daud, Kamilia Kamardin and Nurazean Maarop layers. Although optical buffering implementing fiber delay lines (FDLs) to delay the transmission of bursts is considered to be the most successful technique, but it comes with some drawbacks that is with extra delay and additional expenditure of the FDLs. In this propose work, the FDL is employed between the edge router and the core router; i.e. between IR1, IR2 to CR1 and IR3, IR4 to CR2 as illustrated in Figure 1. The FDL approach that is implemented in this work has the reference value of the study proposed in14. In spite of having a certain length D to delay the light, the propose work has converted it in terms of delaying the data bursts by introducing variable delay time to the link between each edge node to core node so that data burst will have less probability of reaching the core at the same time. The study stated that meters of minimal FDL length can be used in order to obtain a reasonable or optimum blocking rate for a link of 10Gbps. Since in the propose work, the link between IR and CR is set to be 2.5Gbps, therefore the FDL in terms of delaying the burst on the link with respect to time is about 540 ns. Table 3.5 summarize the delay time between each OBS edge node and its corresponding core nodes that are used throughout the simulation works to study the burst loss probability. routed optical networks, owing to the fact that in the former, a burst is sent out without initially reserving resources along the path. Consequently, a burst may be discarded at any intermediate switch down its path, even as it enters its last hop prior to the destination, resulting in considerable waste of network resources. Besides the significant outcome of the wavelength assignment yield, another effective solution to contention resolution technique is the wavelength preemption scheme in providing relative service differentiation. In relative service differentiation, the QoS of one class is classified moderately to other classes. Preemption is the process where high priority traffic has the dominant role to take over the low priority traffic wavelength in this approach of wavelength preemption. By integrating these two substantial techniques, the burst loss probability will definitely be minimized when comparing it with its individual performance in the OBS network. Referring to the proposed core node architecture in Figure 3.8, the traffic classes of EF, AF and BE are assigned each with a predetermined wavelength of 10Gbps capacity. EF is assigned to Lambda 1, Lambda 2 is allocated for AF traffic, whereas BE is being served by Lambda 3. Figure 2. Proposed Core Node Architecture. Figure 1. Proposed FDL Approach. 2.2 Wavelength Assignment and Preemption (WA_Premp) at the Core Router Developments of worthy and resourceful wavelength assignment policies are essential. Such policies are extremely important in OBS networks than in wavelength In this scheme, there is a table for each wavelength to update its bandwidth utilization from time to time as the bursts travel through it. The bandwidth of each link is capped at only 90% of its capacity. When it reaches the capped value, then wavelength preemption technique will take into action and that depends on the traffic priority. In this allocation, Lambda 3 is the lowest priority wavelength. When the allocation bandwidth of Lambda 1 is fully utilized by the EF traffic, the remaining EF burst that need to be sent out will not be dropped but the wavelength of Lambda 3 will be Indian Journal of Science and Technology 3

4 Study of Qos Performance in Optical Burst Switched Networks (OBS) preempted to carry the EF traffic and the consequences BE burst will be dropped. Furthermore, if it comes to worst case scenario, if Lambda 3 is still insufficient to carry the EF traffic, then Lambda 2 will be preempted. In case of the AF traffic, it has the authority to preempt Lambda 3 whereas as for BE traffic, no preemption activity is allowed and therefore, its data bursts are discarded from the network system. 3. Analytical Approach The analytical approach that will be used in this work is the burst blocking probability (BLP) specifically for the performance measure of optical burst switching network. Blocking in a DiffServ Optical Burst Switching network can be modeled with the M/G/k/k queue or mostly known as Erlang B where bursts are not allowed to wait. In this system, k represents the number of wavelengths used at each output port. Evaluation in done based on three classes of traffic. For this analysis, we assume that the basic offset time of each class is the same, which means that the basic offset time does not affect class isolation. The overall burst blocking probability in multi-classes OBS node can be obtained by using the aforementioned Erlang B formula 15. on user satisfaction depending on the application and the abovementioned delay includes delays in the terminal, network and any servers. From a user viewpoint, delay also considers the effect of other network parameters for instance the throughput. Meanwhile, average delay is the average time to transmit each packet from source to destination per flow. In OBS network model, a packet s end-to-end delay comprise of the following four main components and two other sub-components 23 : Processing Delay: It is the time required to sort out a packet in network nodes. It is typically in the order of microseconds or less. Queueing Delay: It is the time as the packets wait to be transmitted onto the link. It depends on the number of other earlier arriving packet that are queue and waiting for transmission across the link. It ranges from milliseconds to microseconds. Transmission Delay: It is the total time needed to transmit all of the packet s bits into the link. It normally ranges from microseconds or less. (2) Where L: length of packets in bits R: Transmission rate of the link in bits/sec Where; P b is the burst blocking probability (BLP) λ is the connection arrival rate µ is the service rate (1) The evaluation of the burst loss probability will be focused on implementation of the propose QoS enhancement in OBS particularly using FDL, FDL_WA and FDL_WA_WPremp with different offered traffic being increased and a different burst assembly schemes employed in the network. Delay apparently can be defined in a several ways, involving the time it consumed to set up a particular service from the first user request and the time to obtain specific information as soon as the service has been set up. Delay has a very strong influence Propagation Delay: Is the amount of time it takes for a certain number of bytes to be transferred over a medium. Propagation delay is the distance between the two routers divided by the propagation speed. Propagation delay is dependent solely on distance and two thirds of the speed of light 24. It is in the order of milliseconds. (3) Where: d: distance between two routers S: propagation speed, which depends on the physical medium of the link. If optical link is being used, therefore the propagation speed would be the speed light. 4 Indian Journal of Science and Technology

5 Suriani Mohd Sam, Salwani Mohd Daud, Kamilia Kamardin and Nurazean Maarop Transmission delay is generally negligible and propagation delay varies with the distance. Therefore, there is not a lot can be done about it since the speed of light is being constant. But, queueing delay is very important. Since the arrival times of these packets at a switch or router are not predictable, this delay is the primary source of jitter. Besides the above-mentioned delay that occurred for a packet transmission in a network, there are also several other sub-components delay that contribute to the overall end-to-end delay. These delays include the delay caused by implementing FDL where burst is delayed by the delaying time. Another sub-component delay is the burst assembling time, which the time is taken for the packets to be assembled to form a burst corresponding to the burst assembly technique used. From the above derivation and other delay contributors, the End-to-End Delay is the delay of a packet from its source to its destination. End-to-End Delay = Q [ D proc + D trans + D prop + D queue ] + D fdl + D BAT (4) where Q: number of router, D proc : Processing Delay, D trans : Transmission Delay, D prop : Propagation Delay, D queue : Queueing Delay, D fdl: Accumulated Delay caused by FDL, D BAT : Burst Assembling Time Delay. All packets in a flow do not experience the same delay in the network. The delay experienced by every packet can differ depending on the transient network conditions. Queues will not occur at the routers if the network is not congested and this contributes to a minimum delay offered by the network. In this work, the average delay is determined by taking the end-to-end delay for each packet divided by the total number of packets as follows: Average delay: (5) when developing the simulation model: IP packets arrivals into the OBS network are based on the traffic characteristics that is CBR for EF traffic, Pareto distribution for AF traffic and Exponential for BE traffic class. Even that is the case, when it reached the core router, it is assumed to be in Poisson. Their arrival rate is denoted by λ and is uniformly distributed over all sender-receiver pairs. IP packet length is varied for different traffic class. Each switch port has 3 wavelengths with a transmission rate of 10 Gbps. Besides that, it is assumed that all nodes support data burst grooming capacity and adopt the priority queueing algorithm to schedule data bursts at the core nodes. Only fixed time burst and hybrid burst are considered in this simulation work and assumed all sub-bursts can be groomed as long as their accumulated length is less than the minimum required length. The DiffServ-aware OBS network model used in this work is shown in Figure 3 is assumed able to cater for three classes of multimedia traffic, i.e. real time constant bit rate (CBR) traffic or Expedited Forwarding (EF) traffic, real-time variable bit rate (rtvbr) traffic or Assured Forwarding (AF) traffic and non-real time variable bit rate (nrtvbr) traffic or Best Effort (BE) traffic. The EF traffic is associated with the uniform traffic pattern and AF traffic is associated with the Pareto traffic pattern. Whereas the BE traffic corresponds to the traffic of the traditional Internet that follows Poisson traffic pattern. The link rate between the sources to the ingress node is set to be at 100Mbps as it will represent the common LAN network in the real environment whereas 25 Mbps of link rate is used between the ingress nodes to the core node. Since this is an optical transmission, the links between the cores are set to be reasonably high at 10Gbps as mostly used in studies. In this topology, it is assumed that each node has no wavelength conversion capability. where X i is the delay of the ith packet n is the number of packets received 4. Simulation Environment The network model of the simulation is shown in Figure In this network, it is assumed each fiber link is unidirectional. The following assumptions were made Figure 3. OBSQoS Provisioning Network Model. Indian Journal of Science and Technology 5

6 Study of Qos Performance in Optical Burst Switched Networks (OBS) The size of the IP packets is variable in size corresponding to its application in the system. The EF traffic is set to have a fixed size of 256 bytes since it is suitable for uncompressed video or voice sources 16, whereas the AF packet size is fixed at 1000 bytes because it is suggested in 17 that packet size for video should not exceed Maximum Transfer Unit (MTU) which is 1500 bytes. Meanwhile, 1500 bytes is the size for the BE traffic class. In this propose network, JET signaling is used whereby in this signaling method, sufficient time is allocated for the data burst to be transmitted without colliding into the BHP. The type of service and the traffic generated together with its traffic rate and packet size is shown in Table As for the burst assembly parameters; the timer burst assembly uses a preset time for every traffic which is illustrated in Table EF has the lowest timer for it since its QoS requirement stringently avoiding delay. Table 1. Type of Service Traffic Description (Packet) Traffic Rate (kbps) Traffic Type & Packet Size EF premium 284 CBR, 256 bytes AF assured 384 Pareto, 1000 bytes BE best effort 256 Poisson, 1500 bytes Table 2. Type of Service Traffic Description (Burst) Fixed Timer Burst (ms) EF premium 4.8 AF assured 55 BE best effort 600 With the aimed to study the average end-to-end delay of the packet for the respective traffic, a simulation is run to observe EF, AF and BE packet average endto-end delay for the fixed timer burst and hybrid burst assembly. The three types of traffic are transmitted separately. The results are illustrated in Figures 4(a), 4(b) and 4(c) for EF, AF and BE traffic respectively to the proposed contention resolution, FDL_WA_WPremp. The end-to-end delay of EF and AF is tens of milliseconds and end-to-end delay of BE is in hundreds of milliseconds. Figure 4(a). EF Packet Average End-to-End Delay Figure 4(b). Comparison of EF Packet Average End-to- End Delay. Figure 4(b) illustrates the average end-to-end delay for an EF packet when it is simulated without the proposed contention resolution, FDL_WA_WPremp. The end-to-end delay it experienced is very much lower in microseconds when compared with the end-to-end delay obtained in Figure 4(a). This is the trade-off when implementing the proposed contention resolution, FDL_ WA_WPremp. It gave a higher delay but minimal burst loss. 5. Performance Evaluation From the ITU-T recommendation of series G , i.e., end user multimedia QoS categories, the performance targets in audio and video application for one-way delay is preferred to be an average of less than 150 ms and 400 ms to the limit. As for data applications, the one-way delay performance target is estimated to be in the range of less than 200 ms to a maximum of 4s depending on the type of data application sent. Figure 4(c). Figure 4(d). AF Packet Average End-to-End Delay BE Packet Average End-to-End Delay End-to-end delay is of concern in real time applications where they need to be avoided or kept at a 6 Indian Journal of Science and Technology

7 Suriani Mohd Sam, Salwani Mohd Daud, Kamilia Kamardin and Nurazean Maarop minimum to ensure smooth performance. From Figure 4(a), Figure 4(c) and Figure 4(d), the BE best effort traffic has the highest average end-to-end delay whilst the EF premium traffic has the lowest. This supports the notion of having lower delay and jitter for real time traffic. Besides that, from the graphs obtained for the three traffic classes, it is observed that by using hybrid burst assembly, the average packet delay is reduced to 21.7%, 17.2% and 13% for the EF, AF and BE traffic class respectively. This can be explained as follows. When a burst is assembled using hybrid scheme, it is not restricted to the order of only one fixed time or fixed size. Instead, the burst will be transmitted when either the time or the size is reached. Thus, the delay incurred would be minimized rather than waiting for the assembled time to reach in fixed time burst assembly. Figure 5(a). EF Bandwidth Utilization versus Offered Load. Figure 5(b). AF Bandwidth Utilization versus Offered Load. Figure 5(c). BE Bandwidth Utilization versus Offered Load. Figure 5(d). Bandwidth Utilization versus Offered Load. Figure 5(a), Figure 5(b) and Figure 5 (c) plots bandwidth utilization versus offered load for the EF, AF and BE traffic class respectively. The graphs are plotted based on the two different burst assembly scheme, i.e., fixed timer burst and hybrid burst for the proposed contention resolution scheme FDL_WA_WPremp. In this simulation works, bandwidth utilization is calculated based on ratio of the traffic load on the link s capacity. The bandwidth utilization from the available traffic starts to rise up significantly until the load is around 0.5 and the value starts to increase minimally from thereof. From Figure 5(a) to Figure 5(c), it is observed that EF traffic utilized higher bandwidth than any other traffic in the network. Furthermore by implementing hybrid burst assembly scheme to the proposed FDL_WA_WPremp gives a significant performance improvement of 15% when compared by deploying fixed timer burst as illustrated in Figure 5(a). Since EF is the highest priority traffic, more burst are sent through the network when hybrid burst assembly is used and this leads to higher bandwidth utilization. The AF traffic utilizes a maximum of 87% of the allocated bandwidth and when hybrid burst assembly is employed in the network when compared to only 75% for the use of fixed timer burst in the FDL_WA_WPremp environment as shown in Figure 5(b). This allows for an average improvement of 10% when deploying hybrid burst assembly. The BE traffic showed the least bandwidth utilization as depicted in Figure 5(c). Although BE burst assembly using hybrid scheme together with FDL_ WA_WPremp performs better than fixed timer burst assembly scheme, but the improvement fares marginally at about 7%. The reason to this phenomenon is due to its state of having lowest priority whereby the total number of packet sent is hindered by the other higher priority traffic such as EF and AF for the time of simulation. Figure 5(d) plots the bandwidth utilization when all three types of traffics are simulated simultaneously as the offered load is varied from 0.1 to 0.9. It is observed that the bandwidth utilization reached its maximum usage at a lower load of 0.7 when compared to a single traffic type being simulated as in Figure 5(a) in which it reached its maximum bandwidth utilization when its load is at 0.9. This is due to the starvation of bandwidth when different types of traffic required different QoS requirement. Indian Journal of Science and Technology 7

8 Study of Qos Performance in Optical Burst Switched Networks (OBS) performance compared with when it employed fixed timer burst scheme with FDL_WA_WPremp as shown in Figure 6(b) and 6(c). 6. Conclusion Figure 6(a). Figure 6(b). Figure 6(c). EF Packet Throughput versus Offered Load. AF Packet Throughput versus Offered Load. BE Packet Throughput versus Offered Load. EF traffic assembled using hybrid burst assembly has the highest througput among the three traffic classes with the ratio of 0.96 when the offered load is increased to 0.9 as illustrated in Figure 6(a). This shows that 96% of its tranmsitted packet were successfully received by its destination and the remainder is loss or dropped during transmission. The significant improvement was compared by the work done in 21 whereby for the same amount of load and similar DiffServ-aware OBS network approach, it obtained only 85% throughput. It is therefore validated that the proposed contention resolution scheme FDL_WA_ WPremp is able to minimize burst loss probability and improved the overall throughput performance by 10%. The same throughput performance pattern goes for the AF and BE traffic with the ratio throughput value of about 93% and 65% respectively as shown in Figure 6(b) and Figure 6(c). The overall performance of AF and BE with the FDL_WA_WPremp technique at the core node fares marginally better when validated with the work done in 21 by only a slight increment of 5% and 2% respectively. The small amount of BE packet throughput is due to its lowest in priority and is dropped due the the preemption contention activities. Both AF and BE traffic implementing hybrid burst assembly with FDL_WA_WPremp showed better throughput This paper describes the QoS performance of the proposed contention resolution technique FDL_WA_ WPremp to the different types of burst assembly, i.e. fixed timer burst and hybrid burst. The QoS performances that are observed include average packet end-to-end delay, traffic class bandwidth utilization and each class throughput. Results show that by implementing hybrid burst assembly scheme with the proposed FDL_WA_ WPremp is able to demonstrate a better performance achievement in terms of average packet end-to-end delay, throughput and bandwidth utilization for all the three traffic classes when compared to employing the fixed timer burst to FDL_WA_WPremp. Besides that, EF marks the best performance achieved as it is the highest priority traffic and meanwhile BE is the least. EF with hybrid burst assembly accomplished 21.7% improvement its end-to-end delay and meets the ITU-T recommendation for its traffic class QoS. Its bandwidth utilization shows a significant enhancement by 15% far better when compared to using fixed timer burst assembly. In addition, the EF traffic class also exhibit almost 95% of the throughput count which is 10% better when compared to the work done by Keping, L., Rodney S.T. and Chonggang, W 25. The AF and BE traffic using hybrid burst assembly with FDL_WA_WPremp also gives better performance compared to using fixed timer based assembly. Although its results are not as significant as showed by the EF traffic, it does contribute to an enhancement. 7. Acknowledgement The authors would like to thank Advanced Informatics School of UniversitiTeknologi Malaysia, Kuala Lumpur and the Ministry of Higher Education of Malaysia. This work was supported in part by a grant from UniverisiTeknologi Malaysia. (Grant No. QK J48). 8 Indian Journal of Science and Technology

9 Suriani Mohd Sam, Salwani Mohd Daud, Kamilia Kamardin and Nurazean Maarop 8. References 1. Toledo MCFD, Zucchi WL. Simulation of a Optical Burst Switch Using Fiber Delay Lines. International Conference on Microwave and Optoelectronics p Li GM, Li VOK, Li CY, Wai PKA. Delayed Reservation Decision in Optical Burst Switching Networks with Optical Buffers. International Conference on Communication and Networking, ChinaCom p Choi JY, Kang M. Optical Buffer-Based Service Differentiation in Burst Switched-Based Optical Networks. International Conference on Advanced Communication Technology, ICACT. 2006; 2: Lee SK, Sriram K, Kim HS, Song JS. Contention-Based Limited Deflection Routing Protocol in Optical Burst Switched Network. IEEE Journal on Selected Areas in Communication. 2005; 23(8): Chan Y, Wu H, Xu D, Qiao C. Performance Analysis of Optical Burst Switched Node with Deflection Routing. International Conference on Communications (ICC). 2003; 2: Long K, Yang X, Hang S, Chen Q, Wang R. An Adaptive Parameter Deflection Routing to Resolve Contention in OBS Networks. International Federation for Information Processing (IFIP). 2006; Zalesky A, Vu HL, Rosberg Z, Wang EWM, Zukerman M. Evaluation of Limited Wavelength Conversion and Deflection Routing as Methods to Reduce Blocking Probability in Optical Burst Switched Networks. IEEE International Conference on Communications. 2004; 3. p Puttasubbappa VS, Perros HG. Performance Analysis of Limited Range Wavelength Conversion in an OBS Switch. Telecommunication System, Springer. 2006; Roseberg Z, Zalesky A, Vu HL, Zukerman M. Analysis of OBS Networks with Limited Wavelength Conversion. IEEE Transaction on Networking. 2006; 14(5): Vokkarane V, Jue JP, Sitaraman S. Burst Segmentation: An Approach for Reducing Packet Loss in Optical Burst Switched Networks. Proceedings IEEE of International Conference on Communications (ICC). 2002; 5. p Detti A, Eramo V, Listanti M. Optical Burst Switching with Burst Drop: An Easy OBS Improvement. Proceedings of IEEE International Conference on Communication (ICC). 2002; 5. p Detti A, Eramo V, Listanti M. Performance Evaluation of a New Technique for IP Support in a WDM Optical Network: Optical Composite Burst Switching (OCBS). Journal Lightwave Technology. 2002; 20(2): Lee YG, Kim NU, Um TW, Kang MH. Decision of the Fiber Delay Line Length in Optical Burst Switching Networks. International Conference on Advanced Communication Technology, ICACT. 2005; 2. p Khlifi Y, Lazzez A, Guemara SEF, Boudriga N. Optical Packet and Burst Switching Node Architecture: Modeling and Performance Analysis. IEEE Telecommunications. 2005; 2: Keshav S, Kalmanek CR, Kanakia H. Rate Controlled Servers for Very High Speed Networks. Global Telecommunications Conference and Exhibition Communications Connecting the Future (GLOBECOM). 1990; 1: de Toledo MCF, Zucchi W. Simulation of Optical Burst Switch using Fiber Delay Lines. International Conference on Microwave and Optoelectronics p Tucker RS. Scalability and energy Consumption of Optical and Electronic Packet Switching. Journal of Lightwave Technology. 2011; 29(16): Available from: K43/K43.pdf. 19. Din NM, Fisal N. Modeling of IP-GMPLS Multimedia Traffic and Network for Admission Control using ns-2. International Conference on Modeling and Simulation, ACMS p ITU-T G.1010, Series G: Transmission Systems and Media, Digital Systems and Networks. Quality of Service and Performance: End-User multimedia QoS Categories Long K, Tucker RS, Wang C. A New Framework and Burst Assembly for IP DiffServ over Optical Burst Switching Networks. GLOBECOM ; [22. Suriani MS, Hamdan S, Norsheila F, Norliza M. Delay Effect on Homogeneous and Heterogeneous Traffic in Optical Burst Switched Network. Third International Conference on Engineering and Applied Science (ICEAS 2013), Osaka, Japan Nov. 23. Kurose JF, Ross KW. Computer Networking: A Top-Down Approach Featuring the Internet. 4 th Edition. Boston: Addison Wesley Lancaster T. Understanding Delay. Unified Communications May. Available from: Understanding-delay 25. Suriani MS, Hamdan S, Norsheila F, Norliza M. Contention Resolution in OBS Using FDL_WA_ WPREMP. International Journal of Future Computer and Communication Feb; 4(1). Indian Journal of Science and Technology 9