Novel Multi-terabit Optical Router Based on Hybrid Switching Technologies

Wei Wei*, Qingji Zeng, Yong Ouyang, Jimin Liu, Xuan Luo, Xuejun Huang R&D Center for Broadband Optical Networking Technologies Dept. of Electronic Engineering, Shanghai Jiaotong University, Shanghai 200030,China
ABSTRACT
Internet backbone network is undergoing a large-scale transformation from the current complex, static and multi-layer electronic-based architecture to the emerging simplified, and dynamic and one-layer photonic-based architecture. The explosive growth in the Internet, multi-media services, and IP router links are demanding the next generation Internet that can accommodate the entire traffic in a cost-effective manner. There is a consensus in current industries that IP over WDM integration technologies will be viable for the next generation of the optical Internet where the simplified flat network architecture can facilitate the networking performance and the networking management. In this paper, we firstly propose a novel node architecture-Terabit Optical Router (TOR) for building the next generation optical Internet and analyses each key function unit of TOR including multi-granularity electrical-optical hybrid switching fabrics, unified control plane unit and so on. Secondly, we give the unified routing definition of multi-layer in TOR and present control plane software structure with emphasis on multi-layer routing issues. Thirdly we describe our cost vs. performance analysis for various application of TOR. According to our calculation, we can get a cost reduction of more than 60 percent by using the TOR. Finally, we reach conclusions that TORs rather than OBS/OPS-based optical routers or big fat router, a cost effective multi-granularity switching and routing technique, are feasible to build the next generation Internet in the coming 5-10 years. IP over WDM, Terabit Optical Router (TOR), GMPLS, Hybrid Switching, Multi-layer Routing.

1. INTRODUCTION
The rapid growth in traffic experienced over the past few years has driven network and service providers to deploy the next generation of high-capacity intelligent backbone optical networks to support IP-centric data and multi-media services. Current estimates predict that Internet backbone traffic will increase by at least 300% per year until 2003, and UUNet, the world's largest ISP, says it expects traffic on its regional trunks to be in the 1-10 petabit range over the next four to five years. As a result, there is a great demand for Gigabit/Terabit routers and witches (high performance IP routers, ATM switches, Ethernet switches, Optical circuit switches, Optical burst switches) that knit together the constituent networks of the global Internet, creating the illusion of a unified whole. From the practical perspective, broadband optical networking converged with IP technologies will be one of best candidates for the next generation Internet backbone network. There is therefore a need to reconsider the evolution of core network, and to depict evolution diagram for all kinds of optical networks today and over the next twenty years as shown in Fig.1.

Correspondence: E-mail: wwei@alloptic.com.cn; Telephone: +86-21-64838770; Fax: +86-21-64518195

174

Optical Transmission Systems and Equipment for WDM Networking, Benjamin B. Dingel, Brewster R. Hemenway, Achyut Kumar Dutta, Ken-Ichi Sato, Editors, Proceedings of SPIE Vol. 4872 (2002) 2002 SPIE 0277-786X/02/$15.00

Fine Packet Switching Coarse Packet Switching Multi-granularity Hybrid Switching Dynamic Circuit Switching Static Circuit Switching
Static Optical Multi-granularity Hybrid Switching Optical Networking OBS/OLS Optical Networking

Pure OPS Optical Networking

Dynamic Optical Networking

Networking

OBS: Optical Burst Switching OLS: Optical Label Switching OPS: Optical Packet Switching

1995-2000 The

1998-2005

2000-2010
The Third

2010-2020
The Fourth Generation

2020-2030
The Fifth

First The Second Generation

Generation

Generation

Generation

Figure 1. The evolution of Optical networking technologies

The current rapid pace of developments in both IP-Centric networks and optical networking are inevitably bringing the two domains closer together [1,2,3,4,5], Solutions to provide the next generation network with high bandwidth, good scalability and easy management are being constantly searched from both IP and optical technology world including OBS-based method [3], Big Fat Router-based method [6] and ASON [2,9]. We think a good, scalable optical Internet architecture, premised on sound architectural principles, is imperative in the quickly evolving Optical Internet environment. Projects in large frameworks of (among others) IST and CA*NET 4 support this research and ITU, IETF, OIF are working hard on the standardizations of the IP over WDM network architecture [9,11]. At present, we are moving toward rapidly achieving the physical devices necessary to build overlay IP over WDM networks architecture, which, however, do not yet meet the requirements of the next generation Optical Internet. In fact, if WDM transmission is clearly expected to meet future Internet traffic capacity requirements, the situation is clearly not settled for multi-layer routing and switching technologies. Hence the efforts to develop a new generation of network architecture, putting forward scalability and flexibility as most critical specifications and trying to avoid complex protocol stacks featuring replication of functionality between different layers. Additionally, aspects and functionalities related to routing and restoration algorithms, control and signaling planes, network and service management will be clearly affected if the architecture and technologies of the next generation networks will not scale adequately in terms of size, performance, and flexibility at the same time. For the above discussion, we present a new paradigm foe the next generation optical Internet architecture based on one-layer backbone as shown Fig.2.

Proc. SPIE Vol. 4872

175

LSR

Aggregation network

Backbone network

Terabit

Optical

Router

OX

Figure 2. The next generation optical Internet architecture base on one-layer

With the standardization of GMPLS by IETF [11], it has become available for carriers to construct such backbone network as Figure 2 based on one layer architecture (from vertical point of view compared to overlay IP over WDM networks)flat, flexible, highly scalable, large-throughput optical network architecture for the delivery of public network IP services. This flat optical Internet architecture helps to lower network cost and maximize revenue. Terabit Optical Routers (TOR) with multi-granularity switching capacity (i.e. wavelength switching capacity, packet switching capacity) based on GMPLS control plane play key roles in Figure 2. TOR in Internet backbone with the ability to simultaneously switch different levels of IP packet granularity inside a box will dramatically improve network performance (i.e. costs, throughout), which will be discussed in section 4, and it can be compatible with legacy node equipment such as IP router, OADM, OXC and so forth. The integration of optical/electrical multi-granularity switch fabrics in the TOR architecture requires modifications and extensions to current IP router or optical switching equipment concepts. This paper is organized as follows. In Section 2, we firstly propose TOR architecture in detail and analyses each function unit of TOR including multi-granularity electrical-optical hybrid switching fabrics, unified control plane and so on. In Section 3, we give the unified routing definition of multi-layer and present control plane software structure with emphasis on multi-layer routing issues. Section 4 describes our cost vs. performance analysis for various application of TOR. After a brief discussion we reach a conclusion that TORs are feasible to build the next generation Internet in the coming 5-10 years.

176

Proc. SPIE Vol. 4872

2. MULTI-GRANULARITY TERABIT OPTICAL ROUTER (TOR) ARCHITECTURE

The Internet has been growing at an exponential rate. Driving this growth is the fact that the Internet has moved from an academy to a mission-critical commercial platform for conducting and succeeding in business. However, the performance of existing network devices (i.e. ATM switch, IP router, and OXC) in these key areas has failed to meet the needs of carriers for those throughput, scalability, reliability and stability issues. Therefore novel nodes for high-capacity and intelligent IP switches are an area of intense research both in telecommunication vendors (i.e. Nortel, Alcatel) and in academic institution communities (i.e. AT&T lab, NTT lab) [2,3,4,6]. The academic and commercial trend towards the R&D of these high-performance routers/switches is based on two aspects: one lies in electrical domain [6]; the other lies in optical domain [2,3]. Fine packet switches are typically designed using electronic technology, and are targeted as an Internet edge router or a MAN/LAN switch. On the other hand, circuit/burst switches/routers are typically designed using optical technology, and are targeted as an Internet core router. At the same time, the push for integrated IP over WDM has given new and strong incentives for other classes of switches/routers such as optical packet/burst switching [3] and hybrid optical electronic switches/routers [4]. Additionally, future prospects for optical multi-granularity switches/routers are an area of intense research [1,4,11]. These multi-granularity optical switches/routers must not only have an aggregate capacity of Gigabits/Terabits per second coupled with forwarding rates of billions of packets per second, but they must also deal with nontrivial issues such as scheduling support for differentiated services, a wide variety of interface types, scalability in terms of capacity and port density, and backward compatibility with a wide range of packet formats and routing protocols.

Novel node?

Hybrid multi-granularity Optical Router

2000

Figure 3. Novel node with the push for integrated IP over WDM

UHWXR5 WD) JL% 567

56*

Electrical domain

2005

UHWXR5 ODFLWS2

ODFLWS2 KFWLZ6

WUDP6

&;2 FLWDW6

Optical domain

Pure OPS -based

2030

Proc. SPIE Vol. 4872

177

In order to build backbones for the massive multi-granularity bandwidth requirements of next generation networks, in our point of view, the integration of IP packet switching with optical switching is a necessity. We believe that the existing optical switching architectures have not yet fully exploited the capabilities of the available devices. Therefore, by using more efficient architectures, combined with more intelligent algorithms and software based on the IP protocol suite (e.g., GMPLS, RSVP-TE, CR-LDP extension, OSPF extension, etc.), we hope to build a new optical router paradigm that can revolutionize data networking. In Figure 4, we proposed a novel architecture design of a GMPLS-based, highly scalable, large-throughput, small-footprint Terabit Optical Router (TOR) with current available components. The proposed architecture integrates packet and wavelength-switching capability.

Unified Management Plane Unit (MPU) Trunk interface (DWDM)

Wavelength Switching Unit (WSU)

Tributary interface (i.e. POS/10GE) Optical Adaptive Unit (OAU) Unified Control Plane Unit (CPU) Packet Switching Unit

Figure 4. The TOR architecture As displayed in the above figure, TOR consists of five units: 1) Wavelength Switching Unit (WSU); 2) Optical Adaptive Unit (OAU); 3) Packet Switching Unit (PSU); 4) Unified Control Plane Unit (CPU) and 5) Unified Management Plane Unit (MPU). Each unit with standard internal communication interface (i.e. Ethernet) conducts its relevant process functions independently. WSU implemented by 3-stage CLOS structure which support 256x256 wavelength switching forwards IP packet in optical domain with coarse granularity; OAU adapts tributary interface data (i.e. OC-192, 10GE) and transport these data to WSU or PSU, it mainly conducts three functionalities-3R regeneration, Wavelength

178

Proc. SPIE Vol. 4872

Converter and lightpath Add/Drop; PSU forwards IP packet in electrical domain with fine granularity; GMPLS-based CPU is the most key component of TOR and process the signaling, routing of in-band or out-of-band (will be described in section 4 in detailed), the CPU that is mainly composed of CR-LDP/RSVP-TE, LMP and OSPF/IS-IS is responsible for all complex functions, including addressing, routing and global control/data topology, resource discovery, traffic engineering, protection/restoration and etc; MPU process all of management information from other units such as alarm information, performance statistical information and configurable information. In TOR architecture, any of the DWDM inputs can be connected to the PSU, where the data streams can be detected and processed by high-level software at the individual packet level, if necessary. The PSU can also perform the function of data aggregation of lower bandwidth streams into a few high-bandwidth channels (i.e. wavelength), which it directs to any output through the WSU. Therefore, WSU and PSU are both essential for building scalable and cost effective TOR that combine the strengths of each technology. Our TOR architecture has some distinct features: (1) multi-granularity switch fabric structure; (2) dynamic traffic/bandwidth grooming capability; (3) intelligent unified control plane based on GMPLS; (4) high-scalability. This scalable and power efficient TOR architecture integrates advanced optical technologies with flexible electrical processing technologies for forwarding packets. This hybrid architecture of TOR base on GMPLS can carry a three-fold effect. First introducing the WSU into a core router node that already uses PSU for wavelength and sub-wavelength switching, we are able to off load the express traffic that inevitably builds up through a junction site. The WSU typically provide the most cost-effective solution to handle express or pass through traffic by keeping that traffic in the optical domain. Second, with this hybrid solution we significantly increase the total switching capacity of junction site allowing more revenue generating services to be managed through the same site. Third, the hybrid solution introduces a new level of restoration flexibility through optical layer [7,8], further adding robustness and efficiencies to the network, while maintaining the necessary functionality of the fine traffic grooming. As today's circuit-switched optical network evolve towards packet-switched optical network, it is important for a networking technology to interoperate with both of them in order to allow seamless network evolution. The optical switching and optical signal Processing technologies coupled with advanced electronics technologies provide wealthy means to create a very intelligent and versatile optical router. In our approach, the IP and optical layers are effectively collapsed into a single layer (or box) that combines the intelligence of IP with the scalable capacity of configurable, multi-wavelength optics.

3. MULTI-LAYER ROUTING ALGORITHM

In section 2 we proposed the TOR architecture [as shown in Figure 4], where TOR has a unified control plane unit (CPU) and a hybrid multi-granularity switching fabric unit (WSU+PSU). The hybrid multi-granularity switching fabric unit can be the ultra-fast forwarding engine of IP packets; the unified control plane unit is used purely for control messages procession with no data forwarding and it is consistent with the Generalized-MPLS (GMPLS) of IETF standardization efforts [11]. GMPLS-based CPU makes it easier to realize the seamless integration of IP networks with optical networks in a box. In particular, integration of the packet switching capability and the wavelength switching technology under a unified GMPLS control plane would significantly enhance the forwarding capacity of the IP network and enable effective network resource utilization of both the IP-layer and the optical-layer in the next generation large capacity IP over WDM network architecture [as shown in Figure 1]. In Figure 1, each TOR would

Proc. SPIE Vol. 4872

179

contain multiple-layer switching capabilities such as PSC and Lambda switching (LSC), LSC and Fiber switching (FSC), etc. These TORs with integrated switch capabilities are required to hold resource information of not only their own link states and network topology but also those of other peer TORs (or the other legacy nodes). In addition, these TORs are also required to discriminate link state information as to which layer these links belong, so as to provide multi-layer routing functionality. The implementation of such functionality enables seamless multi-layer optimized route calculation. The major functional block of the CPU is the desire to implement multi-layer routing protocol, which is responsible for reliably advertising the network topology and resource information (e.g., available link bandwidth) within and between network routing areas (electrical/optical domains). In our TOR architecture, the data plane and control plane topologies is not congruent and the multi-layer routing protocol is responsible for advertising both of these topologies so that each node can maintain a consistent view of the network topologies. The data plane topology is used for path selection during multi-granularity LSP establishment, while the control plane topology is used to build the IP control message routing table. In fact, most of these routing algorithms are the extensions (or constraints) of link state routing protocol such as OSPF, IS-IS and PNNI. We focus on the IETF OSPF [13] based on Dijkstras shortest-path algorithm. Unified Control Plane
Optical layer Data plane Topology DB Electrical layer Data

MIB

Main Control Process

Traffic Engineering Database

LSDB1 LSDB2 LSDB3

Opaque LSAs

plane Topology DB Control plane Topology DB

LSAs

ERO
RSVP-TE CSPF /CR-LDP Algorithm Process Algorithm Dijkstra OSPF Process LMP Process

Path Req.
FIB2 FIB1

Flooding

LSP signaling

Multi-granularity Data Forwarding Plane

Figure 5. The unified control plane software structures of TOR

180

Proc. SPIE Vol. 4872

Fig. 5 depicts the software modules developed for CPU, which include three major functional modules - topology and resource discovery, route/path computation, signaling and restoration. These modules interoperate using standard IPC mechanisms. In the following discussion, we mainly focus on multi-layer routing module. In figure 5, topology and resource discovery of TOR is achieved using an OSPF process. The optical resources interconnecting adjacent nodes are discovered, generally via LMP or OSPF itself. This information is passed to an OSPF process for advertisement throughout the network. Through this process, each node in the network obtains a complete representation of the optical/electrical multi-layer topology and resource utilization. To reduce overheads, in the optical layer, we advertise the aggregated link information in a single OSPF Opaque Link State Advertisement (LSA) rather than advertising each channel separately. An essential part of the routing is to define a common set of attributes to be exchanged over network interfaces. These attributes are independent of the routing protocol used to advertise them. The link characteristics advertised throughout the network are source, destination, link type, link capacity (total, available and reserved channels), a metric defining link preference, switching capacity (e.g., packet, wavelength or fiber switch capable), and shared risk link groups (SRLGs) [10,12,13,14,15]. In order to provide the multi-layer route selection functionality, TOR with Packet Switching Capacity (PSC)+Lambda Switching Capacity (LSC) shall have both layers' link states and network topology data to realize both electrical layer and optical layer route selection. Therefore TOR must discriminate link state data as to which layer they belong. Route selection of Electrical Label Switching Path (ELSP) and Optical Label Switching Path (OLSP) shall be done with reference to link the state data corresponding to each layer (LSDB1or LSDB2 as shown in Fig.5). The link state of an existing OLSP within the multi-layer GMPLS network can be advertised as a point-to-point link (via common control channel) by using conventional router LSA in the case of OSPF. This enables the conventional IP routers, which are connected to the TOR, to realize an IP network plane topology within the network. Alternatively, the link state of existing OLSPs can be advertised in conformance with the concept of FA-LSPs [11]. The OLSP can be identified as the links forming "forwarding adjacency" (FA) between nodes connected by the OLSP. The corresponding LSA of the OLSP (FA-OLSP) shall contain the interface switching capability descriptor, which indicates the OLSP being terminated by PSU, and information about this LSA is used to configure ELSP layer network topology (LSDB2 as shown in Fig. 5). Using the next-hop database created from this network topology data shall do the forwarding of IP packets or ELSPs. The link state of existing fiber-links can be advertised by conforming to the concept of TE extension to routing protocols [13]. The fiber links connected to LSRs are identified as TE-links that forms "routing adjacency" between LSRs. The corresponding LSA of the TE-Fiber Link shall contain the interface switching capability descriptor, which indicates the fiber link being terminated by WSU, and information about this LSA is used to configure OLSP layer network topology (LSDB1 as shown in Fig. 5). Also, these LSRs with integrated PSC+LSC switching capabilities MUST be required to advertise resource information in terms of switching capability. The reservation capability of the integrated switching node capability shall be varied to match the state of resource utilization as discussed in the above discussion. The route of OLSPs can be calculated by the selection of WSU that have OLSP forwarding or termination capability and fiber links that have unreserved wavelength channels. Thus, the TOR can have multi-layer routing capability that enables IP packet forwarding routes to be constructed considering both existing OLSP and fiber-link state. This means that the TOR can realize the IP-centric flexible optical network essential to the next generation terabit rate optical telecommunication networks.

Proc. SPIE Vol. 4872

181

4.

COST VS. PERFORMANCE ANALYSIS OF TOR

TOR Combining the IP layer and optical layers of a network to improve the network efficiency and minimizing the interference between layers has the potential of complicating. TOR with unified traffic engineering capability can be applied to the next generation various network application environment including metro backbone area, national backbone area and so on for its high-performance of packet forwarding capability [16], high-level flexibility with traffic/bandwidth grooming and high scalability. It is also compatible with existing equipment such as router or OXC and collaborates with them to supply OLSP or ELSP services in a pure electrical router network or pure optical switching network with lower connection blocking probability. Figure 6 illustrates the TOR capability of network unified traffic engineering under the control of GMPLS.

Wavelength Switch Unit (WSU)

Optical layer pass through traffic Traffic that need 3R generation or WC Traffic that need L3/L2 grooming

Optical Adaptive Unit (OAU)

ELSP traffic OLSP traffic (Add/Drop)

Packet Switch Unit (PSU)

ELSP traffic

Figure 6. Unified traffic engineering under the control of GMPLS The TOR offers the two traffic transport capabilities: ELSP and OLSP. Some source-destination IP router pairs use PSU to carry their IP traffic (ELSP), while others pass their IP traffic across WSU (OLSP). Note that all ELSP and OLSP are controlled and managed by the same operation paradigmUnified Control Plane Unit (CPU) based on GMPLS. This realizes very simple network management as well as dynamic network control for multiplayer operation. OLSPs are very high performance and cost effective. Because the OLSPs are switched as wavelength paths, no packet-by-packet forwarding and the optical pass-through technique will reduce the volume of OC-192c forwarding by about 1/3~1/2 [4]. Figure 7 shows the results of our detailed study on the cost effectiveness achieved by using the proposed multi-granularity GMPLS-based TOR. The calculation is based on IP traffic demand in the next 2-5 years and state-of-the-art electrical routers, as well as optical switching technology (i.e. MEMS). This integration of PSU with WSU will dramatically reduce network cost (as shown in Figure 7). In most cases, PSU and WSU are both key components of TOR needed for the next generation optical Internet architecture to provide scalability, flexibility and agility in a switching solution. A typical configured TOR with terabit capacity will inherently have a significant portion of the pass through traffic that occupy generally 2/3 of total processed traffic in todays Internet infrastructure traffic pattern.

182

Proc. SPIE Vol. 4872

Figure 7. Network node cost comparisons with the same node throughput According to our calculation, we can get a cost reduction of more than 60 percent by using the TOR in terabit rate national backbone optical telecommunication networks. Though Fig 7 indicates nodal cost, it is possible to extend the analysis to the whole network. Therefore, we conclude that TOR based on multi-granularity switch/routing technologies is one of the cost-effective solutions to design high-speed, flexible, scalable and simple nodes equipments compared with those complicated pure electrical terabit router in the next generation deployable terabit rate optical telecommunication networks.

5. CONCLUSIONS
This paper presented the Terabit Optical Router (TOR) architecture, a cost effective multi-granularity switching and routing technique, which is based on a unified control plane--GMPLS. We also proposed multi-layer integrated routing technique and it helps to reduce optical layer connection blocking probability. This paper also described unified traffic engineering design scheme. The bandwidth granularity of the optical layer is coarse and equal to wavelength bandwidth. On the other hand, the granularity of the IP layer is flexible and well engineered. With this hybrid solution of TOR we significantly increase the total switching capacity of node. In addition, the IP and optical layers are effectively collapsed into a single layer in the TOR architecture that combines the intelligence of IP with the scalable capacity of configurable, multi-wavelength optics, which can reduce todays duplicative network management cost and improve network performance. We expect it is feasible to build the next generation Internet on the basis of TOR in the coming 5-10 years.

Proc. SPIE Vol. 4872

183

REFERENCES
1. 2. 3. 4. 5. 6. 7. 8. 9. N.Ghani, et al. "On IP over WDM integration," IEEE Commun. Mag. P72 Mar. 2000. Greenberg, G. Hjlmtsson and J. Yates, "Smart Routers - Simple Optics, A Network Architecture for IP over WDM," Optical Fiber Commun. Conf., ThU3-2, March 2000. C.Qiao and M.Yoo, Optical Burst Switching-A New Paradigm for an Optical Internet, J.High Speed Network, Special Issue on Optical Networking, vol. 8, no.1, 1999,pp.69-84. Ken-ichi Sato, et al., GMPLS-Based Photonic Multilayer Router (Hikari Router) Architecture: An Overview of Traffic Engineering and Signaling Technology, IEEE Comm. Mag., March 2002, pp. 96 101. Chunsheng Xin, et al., On an IP-Centric Optical Control Plane IEEE Communications magazine, Sep. 2001 pp.88-93. N. McKeown, M. Izzard, A. Mekkittikul The Tiny Tera: A Packet Switch Core IEEE Micro 1997(1-2), 26~33 Ayan Banerjee, et al., Generalized Multiprotocol Label Switching: An Overview of Routing and Management Enhancements, IEEE Comm. Mag., January 2001, pp. 124 128. Ayan Banerjee, et al., Generalized Multiprotocol Label Switching: An Overview of Signaling and Enhancements and Recovery Techniques, IEEE Comm. Mag., July 2001, pp. 144 151. ITU-T G.8080/Y.1304, Architecture and Specifications for Automatically Switched Optical Network. Forum OIF2000.04, IETF Internet Draft, Jan. 2000 11. Mannie, et al., "Generalized Multi-Protocol Label Switching (GMPLS) Architecture", Internet Draft, Work in progress, draft-ietf-ccamp-gmpls-architecture-02.txt, March 2002. 12. K. Kompella , et al., "Routing extensions in support of generalized MPLS" draft-many-ccamp-gmpls-routing-03.txt (work in progress), March 2002. 13. K.Kompella, et al., "OSPF extensions in support of generalized MPLS" draft-ietf-ccamp-ospf-gmpls-extensions-05.txt (work in progress), March 2002. 14. John Strand, et al., Issues For Routing In The Optical Layer, IEEE Comm. Mag., Feb. 2001, pp. 81 87. 15. Z. Zhang et al., Lightpath routing for intelligent optical networks, IEEE Network, July/August, 2001. 16. Jinhan Song, et al., A framework for unified traffic engineering in IP over WDM networks, Optical Networks Magazine, November/December 2001,pp.28-33. 10. S. Chaudhuri, G. Hjlmtsson and J. Yates, "Control of Lightpaths in an Optical Network," Optical Internetworking

184

Proc. SPIE Vol. 4872