Delay analysis of two way reservation schemes in optical burst switching network.

Mohan, E. ; Puttamadappa, C. ; Gandhi, M. Rajiv 等

Introduction

Optical Switching Technique

Wavelength division multiplexing (WDM) appears to be the solution of choice for providing a faster networking infrastructure that can meet the explosive growth of the Internet. Several different technologies have been developed for the transfer of data over WDM, such as wavelength routing (optical circuit switching), optical packet switching, and optical burst switching (OBS) [1]. Wavelength-routed optical networks have already been deployed and, currently, they represent the most promising technology for optical networks. However, wavelength-routed optical networks, which employ circuit switching, may not be the most appropriate technology for the different applications that will use the emerging optical Internet. Optical packet switching is an alternative technology that appears to be the optimum choice. However, at this moment the technology is not mature enough to provide a viable solution. Optical burst switching is a switching technique that occupies the middle of the spectrum between the well-known circuit switching and packet switching paradigms, borrowing ideas from both to deliver a completely new functionality.

Optical Burst Switching

The concept of burst switching first emerged in the context of voice communications in early 1980s. More recently, optical burst switching, has received considerable attention as an alternative to optical packet switching [2]. In essence, optical burst switching considers the optical layer merely as a buffer-less transparent media for applications.

* Granularity: the transmission unit size of optical burst switching is between the optical circuit switching and optical packet switching

* Separation of Control and Data: control information is transmitted on a separate wavelength (or channel)

* One-Way Reservation: Resources are allocated using one-way reservation. That is, a source node does not need to wait for the acknowledgement back from the destination node, before it starts transmitting of the burst.

* Variable Burst Length: The size of burst is variable

* No Optical Buffering: The intermediate node in the optical network does not require optical buffers. Bursts go through the intermediate node without any delay.

One Way Reservation OBS

Many types of OBS paradigms are available for deployment in future all-optical networks. The concept underpinning most types of OBS is as follows. IP packets destined for a common egress router are assembled into data bursts at an ingress router. A control packet precedes each burst by a time offset; the control packet is electronically processed at a sequence of OXCs to reserve a contiguous sequence of wavelength channels, known as a light path or route, for the pending burst. Then, depending on the type of OBS deployed, the pending burst is entirely switched in the optical domain either with or without acknowledgement; one-way reservation, respectively [3]. OBS types using one-way reservation, such as just enough time (JET), just-in-time (JIT) [6] and Horizon, send a burst before it is confirmed that a light path can be reserved, from the ingress router to the egress router. The mean burst blocking probability is considered the most important measure of quality of service (QoS) in OBS types using one-way reservation.

Two Way Reservation OBS

OBS types using two-way reservation, such as wavelength-routed OBS (WR/OBS) [4], delay a burst at the ingress router until an acknowledgement propagates from the egress router to the ingress router, thus confirming a successful light path reservation. These OBS types overcome burst blocking arising from wavelength contention since if the control packet fails to reserve a light path, the burst is electronically buffered at the ingress router and rescheduled for later transmission. However, two-way reservation prolongs the packet delay since the burst must remain idle until an acknowledgement is received. For OBS types using two-way reservation, the most important measures of QoS [7] is queuing delay at ingress routers. WR/OBS is the only OBS type using two-way reservation for which performance evaluation is considered.

WR-OBS

A two-way reservation scheme has also been suggested but it was assumed that bursts are also sent prior to the receipt of an acknowledgment. Recognizing this as a limitation, schemes have been proposed to provide class-of-service differentiation by offset times, i.e., to assign larger offsets for higher priority traffic. This however would have the effect of reducing the burst loss for high priority traffic, at the expense of an increase for lower priority bursts, especially for dynamically varying traffic loads. The result is reduced network capacity for acceptable packet loss rates.

In OBS networks, lower priority bursts experience loss as a penalty. However, given that each burst may contain a large number of transmission control protocol (TCP)--internet protocol (IP) packets or acknowledgment, each lost burst would affect a number of higher layer connections. Thus, in OBS networks, care should be taken to also minimize the loss of lower priority bursts to prevent this.

In this paper, we propose and analyze an alternative OBS network architecture that requires an end-to-end reservation to satisfy specific service criteria such as latency and packet loss rate (PLR) for burst input traffic. This architecture, shown in Fig.2.1 and termed here wavelength-routed optical burst switching (WR-OBS) [8], assumes a fast circuit-switched end-to-end light path assignment with a guaranteed, deterministic delay, and requires an obligatory end-to-end acknowledgment [5].

The packets are electronically aggregated at the network edge into bursts, according to their destination and class of service (CoS), but with timescale of milliseconds, which is a typical forwarding time of IP routers, making the reservation of resources along the path prior to burst transmission feasible. The aggregation time is strictly determined by the performance parameters such as delay at the edge or the required burst size for the network. At an appropriate point during the aggregation cycle, an end-to-end wavelength channel is requested from a network control node for transmission of the burst between edge routers.

[FIGURE 2.1 OMITTED]

Once a free wavelength is found, the aggregated burst is assigned to it and is transmitted into the core network. Its further latency depends only on the propagation delay because buffering operations with associated non-deterministic delays in core nodes are not required. Concentrating all of the processing and buffering within the edge of the network enables a buffer less core network simplifying the design of optical switches or routers/cross connects in the core significantly, which is particularly important for time-critical traffic and cannot be achieved with the currently implemented IP-router [9] infrastructure that provides hop-by-hop forwarding only.

This requires, however, that the bit rate at the input to the buffers at edge routers is sufficiently high to form bursts on a millisecond timescale. Following transmission, the wavelength channel is released and can be reused for subsequent connections. The network core can either be considered as a passive core or as a network of fast-reconfigurable optical routers/cross connects, where the same controller that allocates wavelengths dynamically sets up end-to-end light paths or circuits.

It is assumed that wavelength conversion in core nodes is not required, because, as previously shown, it brings little benefit to wavelength-routed networks with wavelength agility at the network edge. A centralized network management was assumed in this work as a worst-case scenario. A distributed control scheme would be preferred; however, such a scheme relies on synchronization and fast distribution of information on the state of the network.

The aim of this work is to analyze the network performance under which dynamic WR-OBS would bring significant operational advantages and, in particular, in the reuse, utilization of wavelength channels that are set up only for the required burst transmission time (termed wavelength holding time) and, thus, increased over a much simpler but less adaptable quasistatic logically fully meshed WRON. The calculated values for WR-OBS [10] represent an upper bound for the achievable network parameters, namely the edge delay, bandwidth utilization, wavelength reuse, and idle time, and give design rules on the speed requirements for dynamic routing and wavelength assignment algorithms to make a core network in which resources are assigned dynamically practical. The results can be applied to optimize the design rules of future optical network architectures and quantify the operation regimes that best make use of the static or dynamic network architectures.

Network Architecture and Edge Router Model

In this we are going to discuss about network and edge router architecture assumptions, burst aggregation and timing diagrams, modeling of the Impact of Traffic Statistics on Burst Aggregation

The proposed edge router setup is shown schematically in Fig. 3.1, where bursts are aggregated from packets that are electronically presorted according to their destination and CoS and stored in separate queues.

[FIGURE 3.1 OMITTED]

After a time-out signal indicates that packets have to be transmitted to meet application specific latency requirements, a wavelength request is sent to a control node, an acknowledgment is received and the buffer content is dynamically assigned to a free wavelength. If a free wavelength channel is not available, packets are not lost and, instead, are stored in edge-router buffers but could incur additional delay. An edge router with dimensions is considered where is the number of independent traffic inputs, represents the number of CoS, and is the number of destinations.

Burst Aggregation

The variables used to describe the timing of the burst aggregation cycle are as follows. [t.sub.edge] is the maximum delay, i.e., the time the first packet in the buffer spends before the burst is released into the network. [t.sub.prop,sig] is the propagation delay for a control packet sent from the edge router to the network control for wavelength reservation. [t.sub.prop,ack] is the processing time, i.e., time between arrival of control packet and decision on light path and wavelength. [t.sub.prop,ack] is the propagation delay between the sending and receiving edge router for sending the acknowledgment for a wavelength reservation [t.sub.prop,ack] = [t.sub.prop,sig], assuming that the control packets take the same route between the sending edge router and the control node.

T is the wavelength holding time, i.e., the total time for which a wavelength is reserved. [t.sub.prop,net] is the propagation delay for signal traveling from sending to receiving edge router across the core network [t.sub.tran] = [L.sub.burst]/[b.sub.core] is the transmission time of the burst. [t.sub.idle] = [t.sub.prop,ack] + [t.sub.prop,net] the idle time during which the acknowledgment is sent and before the first packet arrives at the receiving edge router.

The burst aggregation cycle can be described as follows. The edge delay [t.sub.edge] is the elapsed time between the time of the arrival of the first bit of the first packet to the buffer queue until the entire burst is released into the network, so that the average queuing delay for all aggregated packets is [t.sub.edge]/2. This holds true, however, only in the case of Poisson arrival processes.

The arriving packets are aggregated in the buffer until triggered either by a threshold indicating potential buffer overflow or a timeout signal for delay-sensitive data. This occurs when the wavelength request-signaling packet is sent to the control node. The propagation delay for this control packet is [t.sub.prop,sig].

It is assumed that the signaling packet contains information on the source and destination edge routers, the CoS and the quantity of data in the buffer, required to estimate the wavelength holding time, defined as the time necessary to empty the buffer and transmit the data between edge routers.

Processing the wavelength request requires time [t.sub.prop], followed by an acknowledgment packet to be returned to the requesting edge router, with an additional delay [t.sub.prop,ack]. Concurrently with the transmission of [t.sub.prop,ack], a wavelength channel is reserved, setting the start of. In parallel, the burst aggregation continues until an acknowledgment from the control node of a confirmed wavelength reservation is received.

In this paper, we assume that the burst assembly terminates at the point the acknowledgment packet from the controller reaches the edge router, although alternative schemes have also been analyzed. This allows the burst aggregation to continue in parallel with the processing of the wavelength request, thus decreasing the overall delay although the final burst size would have to be estimated by monitoring the buffer filling statistics. Packets arriving subsequently to the receipt of the acknowledgment packet are designated to the next burst.

It takes a finite propagation time across the network [t.sub.prop,net] for the first bit to arrive at the destination edge router, so that the reserved wavelength is idle and not used for data transmission for the period [t.sub.idle] = [t.sub.prop,ack] + [t.sub.prop,net] . The time to complete the burst transmission is ttran = [L.sub.burst]/[b.sub.core], so that the wavelength holding time is given by [t.sub.WHT] = [t.sub.idle] + [t.sub.tran]. In principle, the wavelength holding time could be fixed either by the maximum edge delay or by streaming data, in which case would be less predictable but the light path utilization would increase. The maximum deterministic latency or upper bound on the maximum transmission time that packets experience between entering the core network at the source and leaving the destination routers is

Latency max = [t.sub.edge] + [t.sub.prop, net] + [L.sub.burst]/[b.sub.core]

The arrival of the acknowledgment packet from the control node sets the start of the subsequent burst assembly and cycle repeats.

Results and Discussion

In this chapter we are going to present the simulation results for the edge router and core network performance. We will present the analysis of DTWR OBS based on the edge router delay. In this we discussed about the delay that accrued in edge router due to two-way reservation

This time includes an overhead required for light path setup and propagation delays; we refer to it as idle time in the remainder of the paper. In this section, it is assumed that burst sizes increase linearly, equivalent to the case of CBR traffic arriving to the buffer and for which there is no variation in the burst size [L.sub.burst]. Then, for a constant load and CBR traffic, the burst size is proportional to the edge delay and the input bit-rate, so that [L.sub.burst] = [b.sub.in].[t.sub.edge]

[t.sub.wht] = [t.sub.idle] + [L.sub.burst]/[b.sub.core] = 1/A * [t.sub.idle] (4.1)

Where [t.sub.idle] where is the idle time before the burst reaches the destination edge router plus the time for the acknowledgment and A is the core bit-rate to input bit-rate ratio A = [b.sub.core]/[b.sub.in]. For small values of A, the data transmission time can be in the range of tens of milliseconds, so that [t.sub.idle] << [t.sub.wht]. Time [t.sub.idle] starts to affect the service quality when the values of the are comparable to tidle and dominate the wavelength holding time for high core bit rates such as [b.sub.core] = 100Gb/s, as shown in Fig. 4.1 for, [t.sub.idle] = 2,5,10 ms.

[FIGURE 4.1 OMITTED]

Fig. 4.2 shows the effect for [t.sub.idle] = 5ms and a variation of [b.sub.core] from 20 to 100 Gb/s, where, for [b.sub.core] = 20Gb/s, the [t.sub.wht] is significantly longer than for [b.sub.core] = 100Gb/s.

A parameter following from (2) is the bandwidth per wavelength, which indicates the effective bandwidth of a light path used for transmission of data between edge routers

[FIGURE 4.2 OMITTED]

[B.sub.pere.[??]] = [L.sub.burst]/ [t.sub.wht]

[B.sub.pere.[lambda]] = [b.sub.in] * [t.sub.edge]/ [t.sub.idle] + 1/A -- (4.2)

The influence of on is shown in Fig. 4.3 for [b.sub.core] = 100Gb/s and , [t.sub.idle] = 2,5,10 ms. The increase in bandwidth for the identical values is reduced for higher [t.sub.edge] for [t.sub.idle] = 5ms, values remain below 50 Gb/s for 100-Gb/s physical bit rate.

[FIGURE 4.3 OMITTED]

Fig.4.4 shows the effect of bandwidth saturation for [t.sub.idle] = 5ms and core bit rates varying from 20 to 100 Gb/s. The significance of the results is that [B.sub.pere.[lambda]] remains significantly smaller than the optical line rate for [t.sub.edge < 40 ms, especially for high, such as [b.sub.core] 100 Gb/s.

[FIGURE 4.4 OMITTED]

Conclusion

This project describes and analyzes a WR-OBS network that combines the functions of OBS with fast circuit switching by dynamically assigning and releasing wavelength-routed light paths over a buffer less optical core. The potential advantages of this architecture compared to conventional OBS are explicit QoS provisioning and, compared to static WRONs, are in fast adaptation to dynamic traffic changes in optical networks and more efficient utilization of each wavelength channel.

The proposed architecture ensures a deterministic delay for the optical packets through a known, predefined queuing delay at the edge and burst aggregation and the propagation in the core network. Moreover, it guarantees an acknowledgment of the wavelength assignment for QoS-determined provisioning and uses dynamic wavelength routing. Finally in this paper we calculated different parameters in references with edge delay(i.e wave length holding time, band width per wave length).

References

[1] C. Qiao and M. Yoo [2000], 'Choices, features and issues in optical burst Switching,' SPIE Opt. Networks Mag., vol. 1, no. 2, pp. 36-44.

[2] J. S. Turner [1999], 'Terabit burst switching,' J. High Speed Networks, vol. 8, no. 1, pp. 3-16,

[3] K. Dolzer, C. Gauger, J. Spath, and S. Bodamer [, 2001], 'Evaluation of reservationmechanisms for optical burst switching,' AEU Int. J. Electron. Commun., vol. 55, no. 1, pp. 1-8.

[4] M. Duser, E. Kozlovski, R. I. Killey, and P. Bayvel [2000], 'Design trade-offs in optical burst switched networks with dynamic wavelength allocation,'in Proc. Eur. Conf. Optical Communication, vol. 2, Munich, Germany, pp. 23-24.

[5] C. Qiao, [2000] 'Labeled optical burst switching for IP-over-WDM integration,'IEEE Commun. Mag., vol. 38, pp. 104-114,

[6] J. Y. Wei, J. L. Pastor, R. S. Ramamurthy, and Y. Tsai [1999], 'Just-in-time optical burst switching for multiwavelength networks,' in Proc. IFIP Conf. Broadband '99, Hong Kong, China, , pp. 339-352.

[7] M. Yoo, C. Qiao, and S. Dixit [2000], 'QoS performance of optical burst switching in IP-over-WDM networks,' IEEE J. Select. Areas Commun.,vol. 18, pp. 2062-2071,.

[8] Y. Xiong, M. Vandenhoute, and H. C. Cankaya [2000], 'Control architecture in optical burst-switched networks,' IEEE J. Select. Areas Commun., vol.18, pp. 1838-1851,.

[9] S. Verma, H. Chaskar, and R. Ravikanth [2000], 'Optical burst switching: Aviable solution for terabit IP backbone,' IEEE Network, vol.14, no.6,pp.4853

[10] S. Baroni, P. Bayvel, R. J. Gibbens, and S. K. Korotky[May 1999], 'Analysis and design of resilient multifiber wavelength-routed optical transport networks,'J. Lightwave Technol., vol. 17, pp. 743-758.

(1) E. Mohan, (2) C. Puttamadappa, (3) M. Rajiv Gandhi, (4) R. Sathyan and (5) R. Muthuvel

(1) Research Scholar, Vinayaka Missions University, Salem, Tamilnadu, India. E-mail:[email protected]

(2) Professor & Head, Dept of Electronics and Communication Engineering, SJB Institute Of Technology, Bangalore-560060, India.

(3) Lecturer, (5) Student, Dept of Electronics and Communication Engineering, Pallavan College Of Engineering, Thimmasamudram, Kanchipuram-631 502, Tamilnadu, India.

(4) Student, Dept of Computer Science and Engineering, Kanchi Pallavan Engineering College, Kolivakkam, Tamilnadu, India.