On the Design of Control Plane for WDM Burst Switched Networks

Transcription

1 1 On the Design of Control Plane for WDM Burst Switched Networks Yongmei Sun, Tomohiro Hashiguchi, Vu Quang Minh, Xi Wang, Hiroyuki Morikawa, and Tomonori Aoyama Abstract The mismatch between transmission capacity of optical WDM fibers and switching capacity of electronic IP routers has triggered many research activities on optical switching technologies. Among various optical switching paradigms, optical burst switching (OBS) is considered as a promising candidate for the next generation Optical Internet. It brings together the complementary strengths of optics and electronics. This paper presents the design and implementation of control plane for an overlay mode burst-switched optical network testbed. We propose a functional architecture of OBS control plane. It is designed to be programmable, which makes it capable to support various types of protocols and algorithms. The key design issues are discussed in detail, including control signal format, burst assembly, routing, signaling and scheduling with combined contention resolution. Finally, the experimental results are reported. Index Terms Burst scheduling, contention resolution, control plane, optical burst switching (OBS), Optical Internet T I. INTRODUCTION HE rapid growing bandwidth demand and switching speed demand, driven by the Internet, and recent advances in wavelength division multiplexing (WDM) technology has led to a serious mismatch between transmission capacity of optical WDM fibers and switching capacity of electronic IP routers, which has been promoting a network evolution to the next generation IP-centric WDM optical networks, called Optical Internet. In this scenario, the optical technology is expected to play a stronger role not only for transmission but also for switching. Among various optical switching paradigms, optical burst switching (OBS) [1-2] emerged as a promising one because it is more efficient than optical circuit switching and has looser requirements for optical devices than optical packet switching. In burst switched optical networks, a data burst and its control signal are transmitted on separate channels and switched respectively in optical and electronic domains. The physical separation simplifies electronic processing of the control signal The author is with School of Information Science and Technology, The University of Tokyo, 7-3-1, Hongo, Bunkyo-ku, Tokyo, Japan The author is with School of Frontier Sciences, The University of Tokyo, 5-1-5, Kashiwanoha, Kashiwa-shi, Chiba, Japan and provides end-to-end transparent optical path for the burst, thus optimizing optical switching technologies. At the ingress node, multiple client data with the same egress node address are assembled into one burst. Following its control signal with a short delay called offset time, the burst is sent out without waiting for reservation acknowledgement (one-way reservation paradigm). At each intermediate node, the control signal is electronically processed to reserve resources, while the burst is switched directly in optical domain. The offset time allows the intermediate nodes to complete control signal processing and optical switch configuration ahead of burst arrival. At the egress node, the burst is disassembled back into client data. OBS technology has been extensively studied in recent years [3-10]. Besides pure theoretical research, study of prototype and testbed is a crucial and challenging step toward practical OBS networks. We have designed and implemented a general-purpose and flexible OBS network testbed. In this testbed, all of OBS nodes transmit and switch bursts in data plane, while exchange and process control signals in control plane. We have presented the architecture and design issues of its data plane in [11]. In this paper, we focus on the design of its control plane. The control plane is responsible for processing control signals, controlling and managing data plane. It is designed to be programmable, which makes it capable to support various types of protocols and algorithms. The rest of the paper is organized as follows. Section II proposes a functional architecture of OBS control plane. In section III, the key design issues are presented in detail, including control signal format, burst assembly, routing, signaling and scheduling. We report and analyze experimental results in section IV. The paper is concluded in the last section. II. FUNCTIONAL ARCHITECTURE OF OBS CONTROL PLANE A. Architecture of OBS Testbed Various OBS protocols and algorithms with different trade-offs have been studied until now. They need a network environment to be run and evaluated. In order to investigate the feasibility of OBS and evaluate various protocols and algorithms, we have implemented a general-purpose, flexible and feasible OBS network testbed operating in overlay mode. It consists of multiple OBS nodes, which transmit and switch bursts in data plane, and exchange and process control signals in

2 2 control plane. At the ingress node, multiple IP packets with the same egress node address and QoS requirements are assembled into one burst. Then, the burst is routed and forwarded through the OBS network. Finally, it is disassembled back into packets at the egress node. Each node is designed to perform both burst assembly/disassembly and burst forwarding. Therefore, it can act as not only ingress/egress (source/destination) nodes but also intermediate nodes. Furthermore, it supports class of service (CoS) and wavelength selection for burst assembly. Every node supports two remote WDM links connected to other OBS nodes, and four local channels connected to Ethernet. Each WDM link carries four DWDM burst channels (wavelengths) and one shared control channel. For the details, refer to [11]. B. Requirements and Architecture of the Control Plane In OBS networks, as described above, separation of optical data switching and electronic control processing brings the complementary strengths of optics and electronics. To achieve high bandwidth utilization and low burst loss rate, a flexible, intelligent and efficient control plane is desirable for OBS networks. The functions of the control plane will initially be to reserve resource for bursts within OBS networks, and in the long term, to distribute and maintain state information associated with data plane for discovery functions, as well as protection and restoration schemes. This paper focuses on the former one, including burst assembly, routing, signaling and scheduling. We propose a functional architecture of OBS control plane to offer the functions above. In the control plane, OBS nodes are interconnected by control channels. As depicted in the gray part of Fig. 1, the control plane mainly includes five functional modules: routing, scheduling, signaling, assembly and disassembly. The routing module provides forwarding table; the assembly and disassembly modules control the generating and receiving of bursts; the scheduling module chooses an optimal wavelength channel on the desired output optical link; and the signaling module processes control signal and configures switches for bursts according to routing and scheduling information. Fig. 1 Functional architecture of OBS control plane Making a tradeoff between processing speed and implementation flexibility, these functions are developed in software using C language on a Nios embedded processor within Altera s FPGA. III. DESIGN ISSUES OF THE CONTROL PLANE The efficiencies of protocols and algorithms of OBS control plane are related to the formats of the burst and control signal. Considering the popularity of IP networks, asynchronous variable-sized bursts are employed to match the natural characteristics of IP traffic ideally. The control signal carries all of information required by routing, signaling and scheduling schemes. Great care must be taken when designing a signaling protocol and scheduling algorithm for this kind of bursts, to ensure that they can be implemented in real-time by FPGA. An efficient one-way just-enough-time (JET) [2] signaling protocol and a novel scheduling algorithm with combined contention resolution are designed carefully. The detailed design and implementation considerations are described as follow. A. Burst and Control Signal To simplify implementation, a burst is designed to be a simple aggregation of multiple Gigabit Ethernet frames with the same egress node address and CoS attributes. A time-size-based assembly algorithm [12] is adopted to generate asynchronous variable-sized bursts. Precisely, a burst is generated when either the assembly time threshold or the burst size threshold is reached. The detailed architecture and assembly procedure were presented in [11]. Clearly, how to choose assembly time threshold and burst size threshold is a fundamental and crucial issue. In the control plane, taking full advantages of design flexibility allowed by FPGA, these important parameters can be easily adjusted for various experimental studies. To implement the functions of OBS control plane, we define a new control signal format whose information fields are listed in Table 1. Three types of control signal are defined: SETUP, acknowledgment (ACK) and negative ACK (NACK). A SETUP control signal reserves resource for the corresponding burst. An ACK is replied if the egress node receives the burst correctly, otherwise a NACK is sent back by the node which drops the burst. Destination field is used to support a simple destination-based routing, offset time and burst size fields are used for one-way JET signaling protocol, and the other fields are used to support a novel scheduling algorithm with combined contention resolution. TABLE 1: THE INFORMATION FIELDS OF THE CONTROL SIGNAL Field Description Type The type of control signal Wavelength The wavelength carrying burst Destination Egress OBS node address Source Ingress OBS node address Offset time Offset time between control signal and burst Burst size Burst length in time domain CoS Class of service Deflection ID The count of deflection times Survivability TTL in SETUP, failure reason in NACK Option Reserved The control signal can be transmitted in out-of-band channel or in-band subcarrier multiplexing channel. The latter requires fiber delay lines to compensate processing time of the control signal. The former lowers transmission bit rate and O/E/O conversion speed of the control signal. It also makes the control channel sharable for multiple DWDM burst channels, which leads to a better link utilization. A shared out-of-band control

3 3 channel is adopted to connect OBS nodes in the control plane. B. Overlay Routing Mechanism Although there are lots of routing protocols designed for IP networks, it has to be noted that optical technology is essentially an analog rather than digital technology [13]. Therefore, the intrinsic characteristics of optical networks, e.g. transmission impairments, have to be taken into account while calculating the route in OBS networks. As shown in Fig. 1, overlay-based architecture allows independent address structure and routing mechanism in OBS network domain without any change in current IP networks. Several routing mechanisms have been discussed in optical networks, e.g. hierarchical, source-based and step-by-step routing [13]. For step-by-step routing, path selection is invoked at each node to obtain the next link on a path to a destination. It requires less routing information than the former two methods. Within the OBS testbed, a simple destination-based step-by-step routing is employed. However, at the boundary of the OBS testbed, the integration between IP routing and OBS routing also must be considered. We solve this issue by address mapping. At the ingress node, address mapping is performed first, i.e., mapping from IP destination address to OBS egress node address. Then, the destination-based forwarding decision is made in the OBS domain, that is, determining the next hop OBS node and output port based on egress node address. Table 2 gives an example of forwarding table at an ingress node. According to table 2, packets with destination C should be sent to OBS node 3 in the form of bursts. To reach node 3, the assembled bursts are first sent to node 2 through output port 1. It is important to note that only OBS routing is executed at the intermediate nodes. At the egress node, OBS routing and conventional IP routing are both executed. TABLE 2: AN EXAMPLE OF FORWARDING TABLE AT AN INGRESS NODE IP domain OBS domain Destination Egress node Next hop node Output port C B C. One-way Signaling Protocol Because of large round trip time and large buffer requirement on source node, two-way signaling is not suited for OBS networks with high bandwidth-delay product. High efficient one-way signaling is desirable. To date, JET [2] and just-in-time (JIT) [14] are two prevailing one-way signaling protocols. Compared with JIT, JET achieves higher bandwidth utilization and lower burst loss rate with tolerable complexity induced by delayed reservation, i.e., the reservation starts at the expected burst arrival time. JET signaling protocol is adopted in our OBS testbed to achieve high performance. When implementing JET protocol, practical performance of optical switch must be carefully considered. In the testbed, planar lightwave circuit (PLC) switches, with 3ms switching speed, are employed. Hence, a guard time is used to compensate configuration time and switching time of the switches. In other words, reservation starts ahead of the burst arrival with the guard time. The detailed reservation procedure involves four steps: (1) processing control signal; (2) configuring switch matrix ahead of burst arrival with the guard time TG; (3) keeping bandwidth for burst duration time; and (4) automatically releasing bandwidth. Correspondingly, the minimum of offset time TO is denoted as TP N H + TG, where TP is the processing time of control signal at each node and N H is the number of hops along the path. D. Burst Scheduling Algorithm with Combined Contention Resolution From the viewpoint of burst loss rate and bandwidth utilization, burst scheduling algorithm plays a crucial role in OBS networks. As efficient contention resolutions, a simple deflection routing protocol [15] and a smart wavelength assignment algorithm, called PWA [16], have been proposed in our previous works to reduce burst loss rate. Deflection routing approach provides a detouring path for a blocked burst. In PWA, each ingress node keeps a dynamically updated wavelength priority database for each egress node. When a burst is generated, PWA selects the available wavelength with the highest priority to avoid collision proactively as much as possible. In our OBS testbed, the two approaches above are efficiently combined in a carefully designed burst scheduling algorithm. By comprehensively managing scheduling information tables, electronic buffers, and contention resolutions, burst loss rate is decreased and bandwidth utilization is improved. The details of the burst scheduling procedure are described next. Scheduling information is saved in multiple scheduling tables. For an OBS node with M output links and every link carrying N channels, there are M N scheduling tables denoted as CST(m, n) where 1 m M and 1 n N. Each CST(m, n) contains scheduling information of channel n on link m in period of time TT, including start time and end time of each scheduled burst. A scheduler manages CST(m, n) and keeps track of the unscheduled time of each channel. L, C, Ta, TD and TO denote the output link, output channel, arrival time, duration time and offset time of the burst, respectively. In our OBS testbed, the OBS node is fixed to M=3 and N=4. The diagram of burst scheduling procedure is illustrated in Fig. 2. For a local generated burst, the procedure can be divided into the following steps: 1) When a burst is created at time Tc, its L, TD and TO can be obtained. In this case, the earliest possible arrival time to the local optical switch matrix is Ta=Tc+TO. 2) The scheduler looks up CST(L, n) (1 n N) and finds the available outgoing channels within the time period of (Ta, Ta+TD), which is operated in order of channel priority. If one such channel C exists, step 4 is next; otherwise step 3 is next. 3) The scheduler looks for outgoing channels in a near future. Ta is increased by an increment of TI. If Ta+TD lies within TT, step 2 is next; otherwise we assume a scheduling failure and the burst is discarded.

4 4 4) The output channel C is assigned and the scheduling table CST(L, C) is updated. 5) The burst and its control signal are buffered and sent at time Ta and Ta-TO respectively. For a remote burst generated by other OBS nodes, L, C, Ta, TD and TO are obtained by interpreting its control signal. Thus the scheduler looks up CST(L, C) and tries to find if or not the time period of (Ta, Ta+TD) is available. In the case of success, CST(L, C) is updated and the burst is forwarded or received at time Ta; otherwise the burst is deflected or discarded. Fig. 3 The configuration of experimental network As described earlier, bursts are designed to be asynchronous and be of variable length. Fig. 4 shows the observed optical burst on physical layer. The details of ellipse area, including SFD and the beginning of the burst, are enlarged on the top of the figure. From this figure, we can see that idle sequence was inserted during the guard time before burst arrival. A preamble of 7 bytes and a SFD of one byte, denoted as after 8B/10B coding, were used for bit synchronization. The length of bursts varied between 64 bytes and bytes Fig. 2 Burst scheduling procedure IV. EXPERIMENTAL RESULTS We have built the OBS network testbed and experimentally verified the functions of the control plane described above. Furthermore, practical services run on the testbed have been successfully demonstrated on the International Conference on IP + Optical Network (ipop2005) [17]. A. Verification of Protocols and Algorithms This subsection reports the experimental verification of protocols and algorithms of OBS control plane. Fig. 3 shows the experimental configuration. Three OBS nodes were connected by 20km fibers to form a linear network, while three clients were emulated by an Agilent s Router Tester. Client A and B sent IP packets to client C continuously. Node-1 and node-2 assembled IP packets into burst streams called B ac s and B bc s respectively, and sent them using the same four wavelengths. At node-2, B ac s and B bc s were switched to node-3 in time domain and wavelength domain simultaneously. Node-3 disassembled B ac s and B bc s to client C. The basic functions of OBS control plane, including burst assembly/disassembly, routing, signaling and scheduling, were validated by successful receiving at client C. Here, the bit rate of each wavelength was 1.25Gb/s, and the offset time was 13ms. Idle Preamble+SFD Burst Fig. 4 The optical burst Fig. 5 shows forwarding procedure of B ac s at node-2. On receiving the control signal of a B ac at time T, node-2 first performed routing and scheduling based on the content of the control signal. Then, it forwarded the control signal at time T+50µs. Finally, it started to configure switch matrix at time T+3ms and the B ac cut through node-2 at time T+13ms. The result indicates that the average processing time of control signal at one node is TP=50µs, and the switch matrix is configured ahead of the burst arrival with guard time TG=10ms. It is necessary to note that the obtained guard time TG includes not only configuration time and switching time of switch matrix, but also respond time of power equalizer and amplifier. One-way JET protocol and scheduling algorithm were verified successfully. Fig. 5 Forwarding procedure of B acs at node-2

5 5 We also have demonstrated the contention resolutions. In this experiment, two bursts, Bac and Bbc, were generated almost simultaneously at node-1 and node-2. Due to the negligible transmission delay, collision occurred at node-2 in the case of selecting the same wavelength for two bursts. Table 3 shows the results of the contention resolutions in space domain (case 1) and wavelength domain (case 2), which were obtained by analyzing captured control signals arrived/departed at/from node-2. To simplify statement, the control signals of Bac and Bbc are denoted as CPac and CPbc respectively. Note that only departure time was captured for CPbc, and the capture time of the first arrived control signal was used as a reference time. Wavelength ID identified the selected wavelength. From table 3, we see that, in case 1, Bbc was sent to node-3 directly while late arrived Bac was deflected to node-1. In case 2, by assigning different wavelengths to Bac and Bbc, collision was avoided in advance. Case 1 Case 2 TABLE 3: DEMONSTRATION OF CONTENTION RESOLUTIONS Arrival Departure Wavelength Next hop time time ID CPac 0.01ms 0.05ms λ1 Node-1 0ms λ1 Node-3 CPbc CPac 0.01ms 0.04ms λ1 Node-3 0ms λ2 Node-3 CPbc B. Demonstration of Services We have demonstrated online real-time transmission of video stream data on OBS testbed. As shown in Fig. 6, the experimental setup distinguished from that shown in Fig. 3 in that three clients were three personal computers and node-2 was connected with outside Internet. Two real-time video stream services: a TCP-based video on demand (VOD) service between A and B, and a UDP-based live video chat service between A and C using Windows Messenger through outside Internet, were successfully demonstrated simultaneously. The interoperability with IP networks and the possibility of providing TCP/UDP-based latency-sensitive video services through OBS testbed have been confirmed. V. CONCLUSION In this paper, the design issues of an OBS control plane were discussed in detail. Its protocols and algorithms, including burst assembly, one-way JET signaling protocol and scheduling algorithm were experimentally validated. Moreover, the online video services were successfully demonstrated. Further experiments on contention resolutions are underway. The other issues, such as multicast, resource discovery, protection and restoration, are also critical and need to be studied. ACKNOWLEDGMENT This work is supported by National Institute of Information Communications Technology (NiCT). The authors would like to thank NTT Electronics Corporation for valuable discussions and technical assistance. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] Fig. 6 Experimental setup of service demonstration J. S. Turner, Terabit burst switching, J. High Speed Networks, vol. 8, no. 1, pp. 3-16, Jan C. Qiao and M. Yoo, Optical burst switching (OBS) - A new paradigm for an optical internet, J. High Speed Networks, vol. 8, no. 1, pp , Jan L. Xu, H. G. Perros, and G. N. Rouskas, Techniques for optical packet switching and optical burst switching, IEEE Commun. Mag., vol. 39, no. 1, pp , Jan J. White, M. Zukerman, and H. L. Vu, A framework for optical burst switching network design, IEEE Commun. Lett. vol. 6, no. 6, pp , Jun S. Ovadia, C. Maciocco, M. Paniccia, and R. Rajaduray, Photonics burst switching (PBS) architecture for hop and span constrained optical networks, IEEE Commun. Mag., vol. 41, no. 11, pp. s24-s32, Nov J. Li, C. Qiao, and Y. Chen, Recent progress in the scheduling algorithms in optical-burst-switched networks, J. Optical Networking, vol. 3, no. 4, pp , Apr F. Masetti, D. Zriny, D. Verchère, J. Blanton, T. Kim, J. Talley, D. Chiaroni, A. Jourdan, J. Jacquinot, C. Coeurjolly, P. Poignant, M. Renaud, G. Eilenberger, S. Bunse, W. Latenschleager, J. Wolde, U. Bilgak, "Design and implementation of a multi-terabit optical burst/packet router prototype," Proc. OFC 2002, Anaheim, USA, Mar I. Baldine, M. Cassada, A. Bragg, G. Karmous-Edwards, and D. Stevenson, Just-in-time optical burst switching implementation in the ATDnet all-optical networking testbed, Proc. Globecom 2003, San Francisco, USA, Dec A. Sahara, Y. Tsukishima, K. Shimano, M. Koga, K. Mori, Y. Sakai, Y. Ishii, and M. Kawai, Demonstration of optical burst switching network utilizing PLC and MEMS switches with GMPLS control, Proc. ECOC2004, Stockholm, Sweden, Sep K. Kitayama, M. Koga, H. Morikawa, S. Hara, and M. Kawai, "Optical burst switching network testbed in Japan," Proc. OFC2005, Anaheim, USA, Mar Y. Sun, T. Hashiguchi, N. Yoshida, X. Wang, H. Morikawa, and T. Aoyama: "Architecture and design issues of an optical burst switched network testbed," Proc. OECC/COIN2004, Yokohama, Japan, Jul Y. Xiong, M. Vandenhoute, and H. Cankaya, Control architecture in optical burst-switched WDM networks, IEEE JSAC, vol. 18, no. 10, pp , Oct A. Jajszczyk, Control plane for optical networks: the ASON approach, China Commun. Mag., vol. 1, no. 1, pp , Dec I. Baldine, G. Rouskas, H. Perros, and D. Stevenson, JumpStart: a iust-in-time signaling architecture for WDM burst-switched networks, IEEE Commun. Mag., vol.40, no.2, pp , Feb

6 [15] X. Wang, H. Morikawa, and T. Aoyama, Burst optical deflection routing protocol for wavelength routing WDM networks, Optical Networks Mag., vol. 3, no. 6, pp , Nov./Dec [16] X. Wang, H. Morikawa, and T. Aoyama, Priority-based wavelength assignment algorithm for burst switched WDM optical networks, IEICE Trans. Commun., vol. E86-B, no. 5, pp , May [17] Y. Sun, T. Hashiguchi, V. Minh, X. Wang, H. Morikawa, and T. Aoyama: "Optical burst switching testbed," ipop2005 demonstration, Tokyo, Japan, Feb