Optical Burst Switching Obs For Future Internet Computer Science Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Abstract- Wavelength division multiplexing (WDM) technology is rapidly growing due to the huge requirement of bandwidth and extensive use of internet in latest scenario. Though, in country like India, WDM is not widely spread but technical growth shows that in near future we shall also feel the necessity of Optical Burst Switching (OBS) technique to meet the requirements of huge bandwidth. In this paper, we give an introduction to Optical Burst Switching (OBS) and compare it with other existing optical switching paradigms. Basic burst assembly algorithms and their comparisons, effect of optical buffers at the core node will be discussed. Then a brief review of the early work on burst transmission is provided. Algorithms used at an OBS core node for burst scheduling as well as contention resolution strategies are presented next. The various approaches to improve QoS in OBS network are given in detail. The future work in OBS technology to improve QoS i.e. decreasing burst loss percentage, increasing throughput of the network, probable security solutions are given for the researchers. We carried out simulation work on NS-2 simulator for poission traffic and got the expected graphical trends for improvement of OBS network.

Keywords-Wavelength division multiplexing(WDM) ,Optical burst switching(OBS); Scheduling algorithms; Contention resolution ; Optical buffering


For bandwidth hungry applications, WDM technology has the enormous amount of bandwidth available in fiber cable. In a WDM system, each fiber carries multiple communication channels, with each channel operating on a different wavelength. Such an optical transmission system has a potential capacity to provide tera bytes of bandwidth on a single fiber. WDM technology has the capability to provide the bandwidth for the increase in the huge traffic demand of various applications like audio, video and multimedia, which needs the QoS over a network [1].

The currently existing switching techniques can be broadly classified into Optical Circuit Switching (OCS), Optical Packet Switching (OPS) and Optical Burst switching (OBS) techniques [1], [2]. In OCS, an end-to-end optical light path is setup to avoid optical to electronic conversion at intermediate nodes. The main drawback of OCS is circuit setup time and improper holding time of resources. On other hand, no circuit setup is required in OPS but packet header needs to be processed in the electronic domain on hop-by-hop basis, which is very complex and challenging in high speed networks. OBS combines the merits of coarse gained OCS and fined gained OPS [1], [2], [4], [5]. OBS backbone optical networks are playing major role in this sense.

Fig.1. Optical burst switched network

In OBS network model, as shown in Fig.1, there are two kinds of router, edge and core router. The edge router is responsible for deciding the offset time value, generation of Burst Header Packet (BHP), burst assembly and disassembly functions. Core router will forward the burst to its destination node [3], [6]. In OBS, a burst consists of header and payload called data burst. A burst header is called as BHP. The BHP and payload are send separately on different channels. A BHP sends an offset time before the burst transmission. Thus, BHP reserves wavelength for arriving burst in intermediate OBS nodes as shown in Fig.2. Typically, BHP contains information about burst size and burst arrival time. Depending upon this information the wavelength is reserved for the incoming burst for that duration [5], [3], [7], [6].

In OBS network, different wavelength reservation schemes are used for reserving the wavelength. One is called as Tell-And-Wait (TAW), in which when source has the burst to send, it first reserves the wavelength along the route by sending ``request'' message. If the wavelength is granted by intermediate nodes along its route, a positive acknowledgment (PACK) message returns to source from destination; otherwise negative acknowledgment (NACK) is received at source [5], [8], [9], [11].

Fig.2. Separated transmission of data and control signal

Second scheme is called Tell-And-Go (TAG), in which two reservation schemes has been proposed. They are Just-Enough-Time (JET) and Just-In-Time (JIT). In JET, reservation is made by using BHP information. The reservation is made for the duration of data burst. The resources are reserved and released implicitly. In JIT, the resources are reserved as soon as BHP is received and holds these resources until burst departure time. The resources are released explicitly by sending another control message. This results in bad resource utilization. Due to this the wavelength holding time to that node is larger than burst transmission time [5], [8], [9], [12].



Basically, there are three different assembly schemes, threshold-based, timer-based and hybrid-based [10], [13]. An assembly of the burst at the edge router is a challenging issue in the OBS, as it affects the network traffic.

In a timer-based scheme, a timer is started to initialize the assembly process. A burst containing all the packets in the buffer is generated when the timer exceeds the burst assembly period [10]. While in a threshold-based scheme, a burst is created and sent into the OBS network when the total size of the packets in the queue reaches a threshold value [10].

To choose the appropriate time-out or threshold value for creating a burst is still an open issue. A smaller assembly granularity leads to a higher number of bursts and a higher number of contentions, but the average number of packets lost per contention is less and it also increases the number of BHPs. If the reconfiguration period of optical switching fabric is non-negligible, a smaller assembly granularity will lead to lower network utilization because each switched burst needs a reconfiguration period. On the other hand, a higher assembly granularity will lead to a higher burst assembly delay and the average number of packets lost per contention is larger. There is a tradeoff between the number of contentions and the average number of packets lost per contention. The selection of optimal assembly granularity is strongly correlated to the type of input packet traffic [10].

The drawback of the threshold-based scheme is that it does not provide any guarantee on the assembly delay that packets will experience. The drawback of the timer-based scheme is that it does not provide any guarantee on the size of burst.

To overcome the drawback of timer-based and threshold based schemes, hybrid scheme has been proposed. Hybrid assembly scheme is the combination of both threshold-based and timer-based assembly scheme. In the hybrid assembly scheme, a burst can be sent out when either the burst length exceeds the desirable threshold value or the timer expires. In our simulation, we used threshold-based assembly scheme for the burst generation.


After a burst is generated, the burst is buffered in the queue for an offset time before being transmitted to give its BHP enough time to reserve network resource along its route. During this offset time, packets belonging to that queue may continue to arrive. These extra packets are dropped [10]. The burst is preceded in time by a BHP, which is sent on a separate control wavelength. The preceded time is called as "offset time''. The BHP requests resource allocation at each node. At each intermediate node, the BHP is processed electronically and the time taken for processing a BHP is known as the "processing time'' [10], [13], [16].

Various traffic classes exist in OBS. To provide the QoS mechanism in OBS, higher priority classes are suggested to have the higher offset time value. So, the resources will be reserved well before the actual burst arrives. Offset time value decreases with decrease in priority classes [9].

The value of the offset time needs to be decided in such a way that BHP reaches the destination well before arrival of burst. If the offset value is enough, the dropping probability of burst gets reduces, but burst waiting time at the edge node increased simultaneously. If the offset value is not enough, the dropping probability of burst gets increases, but burst waiting time at the edge node decreases simultaneously.

And hence, how to choose the appropriate value of offset time is still an open issue.


Traffic generation is at the heart of every data communication system analysis and design process. A traffic generator that can produce real time and real world traffic proves to be invaluable for analyzing communication systems. Traditionally, most of the simulation in OBS is based on poisson traffic. Poisson traffic uses exponentially distributed inter-arrival times. Poisson model is not very much satisfactory because real network traffic is busty in nature and this model does not support busty nature traffic. Poisson process suffers from large scale traffic analysis [13], [16].

Nowadays, the scale of the Internet enlarges rapidly, and data arrival pattern appears to be chaotic and difficult to model. It has been determined from empirical and theoretical research that the incoming traffic will follow self-similar pattern [16]. With the help of self-similar traffic more realistic results can be obtained because it is based on the self similar nature of the real time traffic. The core idea is to use a behavioral model instead of a mathematical model. Self-similar traffic does not suffer from the lack of large time-scale analysis. Self-similar traffic is based on the idea that we aggregate more and more traffic.


Another important factor which affects the network traffic is scheduling algorithms used for the scheduling of bursts. As the arrival of bursts at OBS node is dynamic, scheduling technique must schedule arrival burst on the available wavelength for the entire duration of burst transmission efficiently and quickly. Scheduling algorithm should be able to process the BHP fast enough before the burst arrives to the node and it should also be able to find proper void for an incoming burst to increase channel bandwidth utilization. Exiting burst scheduling algorithms are given below:

Latest Available Unused Channel (LAUC) algorithm [15]: In LAUC, bursts scheduling is done by selecting the latest available unscheduled data channel for each arriving data burst. In this algorithm, a scheduler keeps track of horizon for each channel, it is the time after which no reservation has been made on that channel. It searches for wavelength from latest available horizon for scheduling burst. It searches the wavelength by using last scheduled burst information on each channel. The scheduler assigns each arriving burst to the channel with minimum void formed by that burst on data channel.

Latest Available Unused Channel with Void Filling (LAUC-VF) algorithm [14], [15]: In LAUC, the voids are created between two data burst assignments on one data channel. This is termed as unused channel capacity. LAUC-VF is variant of LAUC. In this algorithm, a scheduler keeps track of void for each channel, and maintains start and end time of voids for each data channels. LAUC-VF searches for the void such that newly formed void due to new burst is very small compared to other voids.

Best-Fit (BF) algorithm [14]: In this algorithm, a scheduler keeps track of void for each channel. It also maintains start time and end time of voids for each data channels. Scheduler tries to search for a void such that newly created void is the smallest void before and after scheduled bursts.

Minimum Starting Void (Min-SV) algorithm [14], [15]: In this algorithm, a scheduler keeps track of void for each channel. It also maintains start and end time of voids for each data channels. Scheduler tries to search for a void such that newly created voids are the smallest voids after scheduled bursts.

Minimum Ending Void (Min-EV) algorithm [14], [15]: In this algorithm, a scheduler keeps track of void for each channel. It also maintains start and end time of voids for each data channels. Scheduler tries to search for a void such that newly created voids are the smallest voids before scheduled bursts.

BF, Min-SV and Min-EV algorithms are the variants of LAUC-VF algorithm. All the void filling algorithms yields better bandwidth utilization and burst loss rate than LAUC algorithm. But the entire scheduling algorithm has a longer execution time than LAUC algorithm.

Fig. 3 shows, how the different scheduling algorithm behaves when new burst arrives at the node. Table 1 shows the comparisons of different scheduling algorithm [14].

Table 1 summarizes the above discussion using the following notations

W : Number of wavelengths at each output port.

k : Maximum number of data bursts(or reservations) on all channels.

Horizon i : Horizon of the ith data channel.

S{i,j} and E{i,j} : Starting and ending time of jth reservation on channel i.

Fig.3. An example of showing how a new burst is scheduled by different scheduling algorithms

TABLE I. Comparison of scheduling Techniques


Typically, contention resolution in traditional electronic packet switching networks is implemented by storing excess packet in Random Access Memory (RAM) buffers. However, RAM-like optical buffers are not yet available. Currently, optical buffers are constructed from Fiber Delay Line (FDLs) . An FDL is simple a length of fiber and hence offers a fixed delay. Once a packet/burst has entered it, it must emerge after a fixed length of time later. It is impossible to either remove the packet/burst from the FDL earlier or hold it in the longer. The fundamental difficulty facing the designer of an optical packet/burst switch is to implement variable-length buffers from these fixed-length FDLs.

In OBS Contention will occur if multiple bursts from different input ports are destined for the same output port at the same time [10].

Following are some Contention Resolution approaches in Optical Burst switched networks to increase the throughput and to reduce the burst dropping probability.

1) Optical Buffering(Limited Buffering)

2) Wavelength Conversion

3) Deflection Routing

4) Burst segmentation etc.


Optical buffering also known as limited buffering. Current optical buffers may be categorized in different ways. They can be classified as either single-stage, i.e., having only one block of parallel delay line, or multi- stage, which have several blocks of delay lines, cascaded together. Single stage optical buffers are easier to control, but multi-stage implementation may lead to more saving on the amount of hardware used [10].

Optical buffers can also be classified as having feed-forward feedback configuration. Following fig.4 and 5 shows single-stage optical buffer architecture. In a feed-forward configuration, delay lines connect the output of a switching stage of input of the next switching stage. In a feedback configuration, delay lines connect the output of switching stage back to the input of same stage. Long holding time and certain degrees of variable delay can be easily implemented with the feedback configuration by varying the number of loops a packet/burst undergoes. However, each loop causes some loss in single power. Therefore, a packet/burst cannot be stored indefinitely in feedback architecture. In a feed-forward configuration, delay lines with different lengths must be used to achieve variable delays. This architecture attenuates all signals almost equally because every packet/burst passes through the same number of switches. Hybrid combinations of feed-forward and feedback architectures are also possible [10].

Fig.4. Single-stage feed-forward buffer optical buffer architecture

Fig.5. Single-stage feed-backward optical buffer architecture

Based on the position of buffers, packet switches fall into one of three major categories: input, output buffering and shared buffering. In input- buffered switches, a set of buffers is assigned for each input port as shown in fig.6. This configuration has poor performance due to the head-of-line blocking problem. Consequently, it is never proposed for purely optical implementation. In output-buffered switches as shown in fig. 7, a set of buffers is assigned to each output port. Most optical switches emulate output buffering since the delay in each output optical buffer can be determined before the packet/burst enters it. Shared buffering is similar to output buffering except that all output ports share a common pool of buffers.

Fig.6. Input buffer FDLs Architecture


Fig.7. Output buffer FDLs architecture

Due to their hardware-saving characteristics, multi-stage and/or shared-buffered architectures are predominant in optical switch proposals. Fig.4, and 5 shows two single -stage, shared-buffered switch architectures [17] with feed-forward and feedback configurations where N and B are the number of input ports and the number of FDLs, respectively. They both contain FDL pool that is shared among all output ports. In the feed-forward configuration, packet/burst delayed only once, whereas the feedback configuration allows them to be delayed multiple times. Since the FDLs are optical fibers themselves, it is possible for them to hold multiple packets/bursts of different wavelengths simultaneously. However, this comes at the expense of increased complexity in scheduling algorithms. Compare to single - stage buffer architectures, multi-stage counterparts are much more complex. They contain several primitive switching elements connected together by FDLs, usually in a feed-forward configuration. Multi-stage buffers can achieve buffer depth of several thousands. A delay of 1 ms requires over 200km of fiber. Due to limitations of buffers, optical buffering alone as a means of contention resolution may not be effective under high load or busty traffic conditions.

In addition to buffering bursts optically, it is also possible to buffer bursts electronically. Electronic buffering can be accomplished by sending the bursts up to the electronic switching or routing layer. The disadvantage of such an approach is that the network loses transparency, and each node must have electronic switching or routing capabilities, resulting in higher network costs and also requiring electronic memories which must keep up with the speeds of optical networks. Furthermore, a greater load will be placed on the processing capabilities of the electronic switch or router. An alternative would be to implement electronic buffers directly as a part of the optical switch itself. In this case each node will require additional transmitters and receivers, and would need to be aware of the transmission format of the bursts; however no additional electronic routing or switching capability would be required.


In order to evaluate the performance of OBS with burst dropping probability, we developed a simulation model. In our simulation, we consider ring network topology which consists of 24 core and 16 edge nodes with bidirectional links (nsf-topology).

Fig.8. Nsf topology model

Average node degree is 2 and average hop (H) is 5.8. A bidirectional link is realized by two unidirectional links in opposite direction. Each unidirectional link consists of 8 data channel and 1 control channel. Burst arrivals in the network are self-similar or poisson with arrival rate λ. Bursts are generated by using threshold-based assembly schemes and value of threshold is 40 KB. Packet length is kept as 2000 bytes. Shortest path is used for routing the burst from source to destination. Two different classes of traffic are considered namely class 1 (Lower priority) and class 2 (Higher priority). BHP processing time at each node is δ1 µs. Offset time of class 1 traffic is δH. Offset time of class 2 traffic is 10δH which is greater than class 1 traffic. Range of traffic load is from 0.2 to 0.6. It is ensured that BHP is processed well before the data burst is transmitted. Bursts are uniformly distributed over all sender-receiver pairs.

Fig.9. Performance of various algorithms under self-similar traffic

Fig.10 Performance of burst dropping probability of LAUC algorithm under poission traffic with and without FDLs ; per outgoing channel.

Fig.11. Performance of class 1 traffic for various algorithms under poisson traffic using optical fiber delay lines of delay in microseconds.

Fig.12. Performance of burst dropping probability of LAUC algorithms under poisson traffic with and without FDLs.

Fig.13. Performance of throughput of LAUC algorithms under poisson traffic with and without FDLs.

Fig.14. Performance of burst dropping probability of Min-EV algorithm under poission traffic with and without FDLs ; per outgoing channel.

Fig.15.. Performance of throughput of Min-EV algorithms under poisson traffic with and without FDLs.

Fig.10, indicates that Min-EV (Void filling algorithm) has better performance in terms of burst dropping percentage than LAUC algorithm. Initially, the burst dropping percentage in OBS network for poisson traffic without FDLs is calculated then FDLs per outgoing channels are increased from zero to three; the class1 traffic observations are shown in fig.10, for LAUC and Min-EV algorithms. It is observed that in both algorithms burst dropping percentage decreased as FDLs per out-going channels are increased. At the same time, it is observed that burst dropping percentage is less in case of Min-EV algorithm.

Instead of using FDLs per outgoing channels if we will use small fiber delay lines at the core nodes ranging from 5 microseconds to 25 microseconds one at a time as feed-forwarded buffer then we get the amazing results i.e we get less burst dropping percentage by using just one FDL of few microseconds delay at the core node; which is contention resolution approach using optical buffering. Following fig. 11, shows that we get the equivalent dropping percentage with 5 microseconds delay fiber which is generally achieved by two long enough fibers per outgoing channel of hundreds of kilometer in length. We can delay burst by 1ms using fiber of 200 km in length which is not cost effective approach. So, efficient way to reduce the burst loss percentage and to increase the throughput is contention resolution approach. Figs. 12,13,14,15 indicates that burst dropping probability decreased and throughput increased as we used optical buffer at the core nodes.


In this work, under threshold burst assembly scheme with nsf network topology, we study the characteristics of FDLs which affect burst dropping probability. The different traffic like self-similar and poisson is applied to OBS network. The performance of LAUC, and Min-EV is investigated through simulation. As threshold based scheme generate the same size of burst, all void filling algorithm like LAUC-VF, BF, Min-EV and Min-SV have the less burst dropping probability under self-similar and poisson traffic. It is also noticed that as the threshold value increases the burst dropping probability decreases. It has been noticed that the self-similar process gives us the better results with large scale analysis and generate huge amount of traffic but poisson process unable to provide it with same condition. Burst dropping percentage can be minimized and throughput of the network can increased with the help of small fibers at core node of delay in microseconds instead of long fibers per outgoing channel of delay 100 microseconds.

As FDLs per outgoing channels (100 µs each) are increased burst dropping percentage decreased

As Fdls at core node (single Fdl of 05, 10,15…µs etc) is used burst dropping percentage decreased approximately 50% which is generally achieved by use of one FDL of 100 µs in delay, per outgoing channel

The results are analysed for class1 traffic with arrival rate 0.2 and found that during the use of small Fdls of 05 µs delay; the burst dropping probabilty decreases and throughput increases

For self-similar traffic the similar trends of the graphs can be obtained. The ultimate aim is to increase the network throughput using optical buffers at the core node. Hence, Contention resolution with the help of single stage optical buffer; which is one of the parameter to improve the QoS of OBS network is advantageous.

From the present results, our future work is to analyze the effect of Hurst parameter under self-similar traffic as well as Poisson traffic for different types of optical buffers given in this paper with threshold based scheme.