izmailov layout - CiteSeerX

TOPICS IN LIGHTWAVE

Hybrid Hierarchical Optical Networks Rauf Izmailov, Samrat Ganguly, and Ting Wang, NEC USA Yoshihiko Suemura, Yoshiharu Maeno, and Soichiro Araki, NEC Corp.

ABSTRACT Hybrid hierarchical optical cross-connects enhance the performance/cost ratio of optical networks by providing transparent (optical) switching of sets of wavelengths (wavebands) in addition to opaque (electrical) switching of individual wavelengths. As network bandwidth gets cheaper, and the performance bottleneck moves to switching nodes, these systems provide an attractive scalable solution for next-generation optical networks. We describe key technological components (including flexible nonuniform wavebands) of hybrid hierarchical optical cross-connects and discuss their performance/cost implications.

INTRODUCTION The carriers’ networks have been constantly evolving for the last decade following the changing telecom landscape as shaped by market conditions, technological innovations, and regulatory decisions. While historical networks, dominated by voice services, grew at less than 10 percent each year, the explosive growth in data-centric IP services has created demand for capacity that doubles every year. This shift has changed not only the volume of traffic but also its pattern. Whereas the majority of voice traffic is local, IP data traffic remains essentially global. Despite the serious economic slump, the fundamentals of Internet revolution are strong, and the continuing expansion of demand keeps fueling the growth of data traffic. Boosting network capacity is the most obvious strategy. However, carriers realize that the well publicized and widely misunderstood “fiber glut” is not the long-term solution. While building trenches and laying multiple fibers helps to reduce the transportation cost, the key cost components and underlying complexity are being shifted to the bottleneck switching and regeneration nodes. In order to satisfy the growing bandwidth demands, more diverse and intelligent allocation of capacity is required. Optical networking has become a key technology to accommodate the rapidly expanding Internet traffic. New networks are expected to support the increasing network load by employing advanced transmission (wavelength-division multiplexing, WDM) and switching (optical switches and cross-connects, OXCs) technologies

88

0163-6804/02/$17.00 © 2002 IEEE

[1]. After witnessing several disruptive technological waves over the last decade, carriers want to keep their options open and leverage the value of their investment in a way that takes into account both tomorrow’s market demand and foreseeable technological trends. While the reduction of capital and operational expenditure is the most pressing task of the moment, carriers are still looking for a new generation of intelligent optical nodes that are modular, scalable, simple to integrate, and amenable to network solutions providing efficient and flexible services. Hybrid hierarchical OXC has the potential to become the key element of the comprehensive long-term solution that will enable carriers to create, maintain, and evolve scalable and profitable networks. This potential is based on significant capital expenditure savings delivered by hybrid technology, which replaces much expensive opto-electronic fabric with an all-optical one. The potential is augmented by hierarchical technology, which further reduces capital expenditure since the same optical port can process multiple wavelengths simultaneously. Finally, the flexible structure of nonuniform wavebands reduces capital expenditure even more by improving the throughput of the node. In the next section, we outline the main optical technologies that enable hybrid hierarchical optical nodes. Next, we describe the performance of the resulting hybrid hierarchical optical networks. Finally, we address the scope of future work that needs to be investigated in this area.

HYBRID HIERARCHICAL OPTICAL NODES HYBRID TECHNOLOGY BACKGROUND The revolutionary WDM technology increases the transmission capacity of fiber links by two to three orders of magnitude. This huge increase in transmission capacity comes at the expense of managing the wavelengths (separating, combining, adding, dropping, routing, switching, and converting), which has to be handled by the switching equipments. Emerging hybrid hierarchical OXCs may become an attractive solution in next-generation optical networks. On the node level, these systems combine two promising technologies — a hybrid of all-optical

IEEE Communications Magazine • November 2002

and optical-electrical-optical (OEO) cross-connects and hierarchical processing of wavelengths aggregated into wavebands — with flexible nonuniform wavebands. On the network level, they provide for diverse provisioning and protection of various types of traffic. In the next subsections, these key components are described in more detail. Traditionally, two types of switching systems have been deployed in various networks: circuit switching and packet (cell) switching. Electronic networks started as circuit-switched networks (based on ubiquitous telephone connectivity) and then evolved into more efficient packet-based networks more suitable for bursty data traffic. The functionality of circuits (being useful for traffic engineering, quality of service, and network management) was first supported by asynchronous transfer mode (ATM) and later by multiprotocol label switching (MPLS) technologies. Optical networks will likely proceed along the same path as electronic networks. Currently, optical networks are at the stage of circuit switching. An optical packet switching solution, explored for several years already, is not yet a practical or comprehensive solution and is still years away. The circuits in optical networks are being handled by optical add-drop multiplexers (OADMs) and OXCs. These critical network elements sit at junction points in optical backbones and enable carriers to string together wavelengths to provide end-to-end connections. Over the last few years, equipment vendors have been driven by two major goals: automating the provisioning process and boosting the switching capacities to handle the growing number of wavelengths. While pursuing these goals, the vendors have been developing two distinct competing versions of OXCs: OEO (opaque) and all-optical (transparent).

THE OEO-BASED XC An OEO-based XC converts the incoming optical signal into an electrical signal for subsequent switching and grooming; the electrical signal is regenerated as a new optical signal at the output port. OEO-based XCs are based on mature and reliable technology (O/E, digital cross-connect, E/O). They can provide a variety of functions such as optical regeneration, reshaping, and retiming (3R), grooming, and wavelength conversion. Each wavelength is thus processed individually, which becomes expensive as traffic volume grows. An OEO-based XC can be controlled with a distributed control plane. Systems from different vendors can interoperate, and standards are already in place to extend interoperability to the control plane. However, these rich functionalities need to be supported by expensive hardware, which, being protocol-dependent and bit-ratedependent, cannot scale very well. OEO-based XCs are also characterized by large footprint and large power consumption, and generate significant amount of heat.

THE ALL-OPTICAL XC The all-optical XC transparently switches the incoming optical signal through the switching fabric; the optical signal remains in the optical domain when it emerges from the switching fabric. A variety of technologies can be used for

IEEE Communications Magazine • November 2002

these purposes: arrays of tiny tilting mirrors (micro-electromechanical systems, MEMS), liquid crystals, bubbles, holograms, and thermoand acousto-optics. All-optical XCs are less expensive than OEObased XCs, have a smaller footprint, consume less power, and generate less heat. All-optical XCs are highly scalable: multiple wavelengths can be handled using the same port. Being naturally protocol- and bit-rate-independent, they are able to carry services in their native format, providing a future-proof alternative for OEO-based systems. However, the absence of optical 3R functions (required to clean up accumulated optical impairments) and wavelength conversion (required to resolve wavelength contention) restricts the capabilities of all-optical XCs. The lack of interoperability limits the deployment of all-optical XCs as well. Transparency of optical signals also makes it harder to monitor performance in all-optical XCs.

HYBRID TECHNOLOGY Hybrid technology (employing both types of cross-connects, OEO and all-optical) is emerging as an inevitable convergence point leveraging the advantages of both technologies (Fig. 1). With optical networks dominated by rapidly growing long-distance data traffic, most of the traffic passing through the optical nodes (as much as 75 percent) consists of transit traffic. A hybrid cross-connect can use its all-optical part to route transit traffic while the OEO part is used for grooming and adding/dropping local traffic. Electronics is thus used to perform necessary and expensive processing of traffic, while optics is used for inexpensive and transparent forwarding.

A hybrid crossconnect can use its all-optical part to route transit traffic while the OEO part is used for grooming and adding/dropping local traffic. Electronics is thus used to perform necessary and expensive processing of the traffic, while optics is used for inexpensive and transparent forwarding.

HIERARCHICAL TECHNOLOGY BACKGROUND In traditional networks, performance and scalability concerns called for layered mechanisms providing various levels of traffic aggregation, such as differentiated services (DiffServ) and MPLS standards for IP networks. Scalability concerns for packet switching networks called for layered mechanisms providing various levels of traffic aggregation supported by DiffServ and MPLS standards. Similarly, circuit-based carrier networks are hierarchical and are based on a synchronous optical network and synchronous digital hierarchy (SONET/SDH). In case of the optical networking, the same cost and scalability concerns translate into creation of multiple switching granularities, such as wavelengths and wavebands [2–6]. Standard routing protocols cause many traffic flows run alongside each other over the same path of links. This creates an opportunity to aggregate express optical wavelengths into wavebands (also called fat pipes or super channels) that can be optically (transparently) switched for most part of their path through the network without being demultiplexed into separate wavelengths at every node. The optical paths thus form a hierarchy in which higher layer paths (waveband) consist of several segments of lower layer paths (wavelengths). The resulting hierarchy of wavelengths and wavebands is a mixed one: logical wavebands coexist on the same fiber with individual wavelengths, providing better efficiency. The poten-

89

Demux

Treating multiple wavelengths

Mux B1

B1 MK × MK

within a waveband as a

λ1, ..., λN

1

the optical switching

BK

Demux

reduces the size

λ1 ,..., λN

1

BK

single entity and complexity of

Fiber 1

Fiber 1

B1

B1 Fiber 2 λ1, ..., λN BK

2

2

M

M

Mux Fiber 2

λ1, ..., λN

BK

matrices. In addition, optical

OOO

amplifiers can Demux

operate on an entire waveband

Deselector

Mux

Selector B1

B1

Fiber M

Fiber M

without any knowledge of the individual wavelengths.

λ1, ..., λN

λ1, ..., λN

BK

BK Add

Drop OEO

■ Figure 1. A hybrid hierarchical cross-connect node.

tial cost benefits of wavelength aggregation into wavebands was demonstrated in literature [7–10].

UNIFORM WAVEBANDS The hierarchical technology is emerging as one of the key solutions for looming scalability problem of optical networks. All-optical XCs (or the alloptical part of hybrid XCs) can switch large wavebands rather than individual wavelengths, without any extra cost (as opposed to OEO-based XCs that are wavelength-dependent). A waveband path occupies only two (input and output) ports of an optical switch in a node. The path hierarchy reduces node costs since a waveband can be switched optically as a single entity, thus reducing the number of expensive OEO ports required for processing individual wavelengths. Treating multiple wavelengths within a waveband as a single entity reduces the size and complexity of the optical switching matrices. In addition, optical amplifiers can operate on an entire waveband without any knowledge of the individual wavelengths. The hierarchical OXC also cuts carriers’ costs in WDM multiplexers and demultiplexers (mux and demux) and transmitreceive subsystems, which are used before the inputs and after the outputs of the cross-connect. To quantify the advantages of hierarchical technology, we assume that some of the wavelengths can be aggregated into wavebands consisting of G wavelengths each. Each waveband consists of contiguous wavelengths, that is, the waveband number m contains all the wavelengths with numbers from (m – 1)G + 1 to mG. We define the cost of rout-

90

ing in a hierarchical optical network as the total cost of the ports (both OEO and optical ones) required for routing the traffic flows (we assume an optical port costs five times less than an OEO port, which is the unit of cost). Thus, the cost of a wavelength segment consisting of N links is equal to 2G(N + 1), while the cost of a waveband segment aggregating this wavelength segment is equal to 4G + 2(N + 1)/5. Figure 2 illustrates the cost computation for a wavelength segment consisting of N = 5 links and granularity G = 4. Cost-efficient implementation of optical hierarchy has to be delivered by appropriately designed routing and aggregation algorithms. Routing and wavelength assignment algorithms were extensively studied in the general context of optical networking [1]. A hierarchical approach adds another layer of routing, wavelength, and waveband assignment algorithms.

NONUNIFORM WAVEBANDS Aggregation of wavelengths into uniform wavebands (each comprising G wavelengths) introduces the aggregation overhead adversely affecting the hierarchical node’s performance. Consider an optical switching node with M output fibers and suppose that the input fiber carries N wavelengths to be switched to any of M outputs. Depending on the breakdown of N input wavelengths among M output ports (i.e., the set of the numbers of wavelengths switched to each output ports), the packing efficiency of their aggregation into wavebands may vary. Consider the example in Fig. 3. It shows an input fiber carrying N = 8

IEEE Communications Magazine • November 2002

In general, depending on the cost assumptions, mechanisms of Path cost = 2G(N + 1)

Wavelength path

filling the wavebands, and so on, the packing improvement provided by non-uniform

Path cost = 4G + 2(N + 1)/5

wavebands can

Waveband path

be anywhere N = 5 links

G=4

between 20 percent and 40 percent.

Wavelength switch (OEO)

Waveband switch (OOO)

OEO port

OOO port

■ Figure 2. Routing metric: wavelengths and wavebands. wavelengths that have to be switched into M = 4 output fibers (shown as four pipes in the right side of Fig. 3). The numbers of wavelengths to be switched to the four output fibers are equal to (3,1,2,2). In Fig. 3, the wavelengths to be switched to the same output fiber are painted in the same color as that of the output fiber; for example, three red wavelengths are to be switched to the red output, one blue wavelength is to be switched to the blue output, and so on. We assume that the waveband granularity G = 2, the wavelengths can be aggregated into preconfigured uniform wavebands of the size of two wavelengths each. In this example, two switching solutions can be employed. In the first approach (the OEO solution, shown in the upper part of Fig. 3), two expensive OEO ports are used to switch two of the wavelengths in the OEO layer, while three wavebands are used to switch the remaining wavelengths. In the second approach (the all-optical solution, shown in the lower part of Fig. 3), five wavebands are used to switch all the traffic in the all-optical domain. However, the same wavelength demand (3,1,2,2) could have been switched optically if the wavebands had been preconfigured in the way shown as the nonuniform solution (shown in the middle part of Fig. 3): two wavebands containing two wavelengths each, one waveband containing three wavelengths, and one waveband containing one wavelength. Figure 4 illustrates all breakdowns of N = 8 wavelengths between M = 2 output fibers and the equipment costs required to carry those wavelengths (for uniform and nonuniform wavebands). Depending on the particular approach, the uniform wavebands approach requires up to five wavebands (all-optical solution) or three wavebands and two OEO ports (OEO solution)

IEEE Communications Magazine • November 2002

to carry all possible traffic loads, whereas the approach based on nonuniform wavebands consistently requires only four wavebands for all traffic distributions. In general, depending on the cost assumptions, mechanisms to fill the wavebands, and so on, the packing improvement provided by nonuniform wavebands can be anywhere between 20 and 40 percent.

WAVEBAND SELECTION AND REALIZATION The concept of nonuniform wavebands gives rise to the following two issues. The first, the waveband selection problem, is how to preconfigure a set of wavebands that can be used to represent an arbitrary breakdown of input flow of N wavelengths into M output fibers. Thus, the set of waveband sizes should not depend on the traffic pattern. The second, the waveband assignment problem, is how to assign these preconfigured wavebands for optical switching of N wavelengths into M output fibers. In the absence of other constraints, both the waveband selection and waveband assignment problems can be solved in polynomial time by using the following algorithm [7]: • Input N and M and create the empty set B • Assign N* as the smallest integer number larger than N/M • Add the element N* to the set B • Assign N = N -– N* • If N = 0, stop. Else go to step 2. Given the set of optimal wavebands, an arbitrary breakdown of demand of N wavelengths into M output fibers can be realized by sequentially assigning the largest available waveband to the largest remaining wavelength demand. Nonuniform wavebands can be realized by a waveband (de)aggregator subsystem that can realize and process nonuniform wavebands. Tradition-

91

2

3 uniform wavebands and 2 OEO ports (OEO solution)

While uniform wavebands are analogous to traditional mechanisms used in packet- and circuitswitched networks (with their advantages of simplified forwarding and drawbacks of aggregation overhead), the nonuniform waveband framework is a unique optical solution that expands the flexibility of wavelength aggregation and creates attractive cost-efficient opportunities. While uniform wavebands can be compared to one-size-fitsall fat pipes, nonuniform wavebands are in fact diverse pipes that can be manipulated and assigned in a way that creates a system of appropriately sized all-optical shortcuts through the network with minimum overhead usually associated with any type of aggregation. Diverse nonuniform wavebands open a new way of handling the all-optical layer of the network that is essentially future-proof.

M outputs (M = 4)

2

N wavelengths (N=8)

2

Input fiber

3

4 nonuniform wavebands

1 2 2

2

2

HYBRID HIERARCHICAL OPTICAL NETWORKS

2

5 uniform wavebands (all-optical solution)

2

NETWORK APPLICATIONS

2

Hybrid hierarchical optical nodes have the potential to provide various benefits to optical networks. The hybrid technology promises significant capital expenditure savings by replacing a large portion of the expensive OEO part with optical (along with reducing power consumption and footprint). The hierarchical technology further reduces capital expenditure since the same optical port can process multiple wavelengths simultaneously. The nonuniform wavebands reduce the capital expenditure even further by improving the optical throughput of the node, and provide a diverse and flexible system of waveband paths. The all-optical part of the hybrid, being bit-ratefree and protocol-free, provides operational expenditure savings. Wavebands can be created and filled up with wavelengths in service, thus

■ Figure 3. Examples of uniform and nonuniform wavebands. al WDM muxes and demuxes combine (couple) and separate (split) different optical wavelengths. A WDM demux separates the set of wavelengths on an incoming fiber into a number of wavelength subsets. These wavelength subsets can be uniform or nonuniform fixed groups (wavebands), shown in Fig. 5. A demux subsystem may produce both fixed and arbitrary wavelength subsets. Nonuniform waveband deaggregators can be used instead of demuxes in hybrid hierarchical OXCs (shown in Fig. 1), further reducing the cost of the OXC node by employing a flexible set of wavebands containing different numbers of wavelengths.

N wavelengths (N = 8)

M outputs (M =2)

0

1 2 3 4 5 6 7 Number of wavelengths to 1st output fiber

8

8

7 6 5 4 3 2 1 Number of wavelengths to 2nd output fiber

0

1st

Input fiber

Wavelengths directed to First fiber 0 1 2 3 4 5 6 7 8

Uniform wavebands 2λ+2λ+2λ+2λ

Second fiber 8 7 6 5 4 3 2 1 0

Non-uniform wavebands 1λ+1λ+2λ+4λ

All-optical cost 0 2 2 2+2 2+2 2+2+2 2+2+2 2+2+2+2 2+2+2+2

2+2+2+2 2+2+2+2 2+2+2 2+2+2 2+2 2+2 2 2 0

4wb 5wb 4wb 5wb 4wb 5wb 4wb 5wb 4wb

2nd

OEO cost

4wb 3wb+2oeo 4wb 3wb+2oeo 4wb 3wb+2oeo 4wb 3wb+2oeo 4wb

Cost 0 1 2 1+2 4 1+4 1+1+4 1+2+4 1+1+2+4

1+1+2+4 1+2+4 1+1+4 1+4 1+1+2 1+2 2 1 0

4wb 4wb 4wb 4wb 4wb 4wb 4wb 4wb 4wb

■ Figure 4. Switching costs of uniform and nonuniform wavebands.

92

IEEE Communications Magazine • November 2002

simplifying network management, providing seamless upgrade, and further reducing the operational expenditures. The all-optical part of the hybrid also provides necessary future proofing, while the electronic part delivers a seamless migration path to all-optical networks. Hybrid hierarchical architecture also appears to be a promising solution for core metro networks. While WDM has been slow to penetrate the metro market due to greater cost sensitivity (large number of systems per individual user), greater need for protocol transparency (multiple protocols are being deployed), and greater need for flexibility (traffic variability is higher in metro networks), these very problems can be addressed and solved by employing the hybrid hierarchical approach. As WDM systems move from long-haul core to metro networks, the ability to handle large numbers of wavelengths with relatively small numbers of physical fiber ports becomes especially important for metro area applications.

CONTROL PLANE The nodes in the network can be controlled by a centralized controller (centralized control plane) or by their own controllers (distributed control plane, e.g., generalized MPLS, GMPLS) [10]. The distributed control plane performs neighbor discovery, topology discovery, and route computation. The source node of a path may compute the route by using the local image of network resources. The centralized control plane performs the same topology discovery and routing computation as the distributed control plane. It provides better routing performance and network utilization. Depending on the number of hierarchical nodes in a network, we can classify the network as homogeneous or heterogeneous. In the homogeneous model, all nodes are hybrid hierarchical, so a waveband path can be set up between any two nodes. In the heterogeneous model, some of the nodes are hierarchical while others are not. Since the latter nodes do not have waveband switches, a waveband path can only be set up between two hierarchical nodes. We use the sequence of routing and waveband aggregation to classify the routing models as integrated or separate. Under the integrated routing model, the routes are computed for both the wavelength and waveband paths. The routing decision is based on the status of the advertised wavelength and waveband resources. Integrated routing can achieve more cost-effective use of resources, although it uses a complicated routing algorithm (of polynomial complexity). Under the separate routing model, there are two independent modules acting on the wavelength and waveband layers, respectively. First, the wavelength module computes the routes for the traffic demands taking into account wavelength resources. Next, the waveband module determines the possibility of wavelength aggregation into wavebands taking into account waveband resources. If needed, wavelengths may be reassigned to create a new waveband path between intermediate nodes [8].

SIMULATIONS To illustrate the advantages of hybrid hierarchical networks, we assume the following algorithm based on the separate routing model. First, all

IEEE Communications Magazine • November 2002

Uniform deaggregator

Nonuniform deaggregator λ1-80

λ1-40

λ1-160

λ41-80

λ1-160

λ81-120

λ81-120

λ121-152

λ121-160

λ153-160

■ Figure 5. Waveband deaggregators; uniform and nonuniform band sizes. the wavelength paths are computed without any waveband aggregation. Then, starting from a large value MaxHops (equal to the number of hops in the longest path), the algorithm tries to locate a common segment of length MaxHops belonging to G wavelength paths. If such a segment is found, the corresponding waveband is created, and the search is continued until no such segments are found. In the subsequent steps, the parameter MaxHops is decreased by 1 and the procedure is repeated until MaxHops reaches the minimum value of 2. We simulated waveband granularities from G = 2 to G = 20. We assumed that each link of the network could carry 160 wavelengths. The wavelengths could be aggregated into wavebands consisting of G wavelengths each. Therefore, the granularity G determines the maximum number 160/G of wavebands on a link. The network is assumed to be homogeneous: every node is able to perform waveband aggregation. The traffic matrices were created to reflect a skewed traffic distribution as observed in real networks (e.g., in distribution of Web traffic among popular sites): the nodes in the network were randomly ranked and the amount of traffic generated by a node was inversely proportional to its rank (Zipf distribution). All wavelength and waveband paths were assumed to be unidirectional. In order to focus only on the performance of waveband aggregation irrespective of the wavelength assignment policies, we assumed that all the nodes have infinite numbers of ports. We also assumed that the nodes have full wavelength conversion capabilities at the OEO layer. The simulation results for an American optical network are shown in Fig. 6. Waveband granularity G is only defined for uniform wavebands (each waveband has the same size G). All other cases (OEO-based network, all-optical wavelength-based network, nonuniform wavebandbased network) do not depend on G. These results demonstrate significant cost savings that could be achieved by deploying different layers of hybrid hierarchical technology. Specifically, we note that waveband granularity has an impact on the performance of using uniform wavebands. In the scenario studied, waveband granularity G = 4 proves optimal. Furthermore, the results show additional cost reduction employing nonuniform wavebands. Simulations for other networks and traffic patterns support these qualitative observations and can be found in [7, 8].

93

100%

OEO-based network

90% 80%

All-optical wavelength network Hybrid (uniform wavebands) network Hybrid (nonuniform wavebands) network

70% 60% 50% 40% 30% 20% 10%

BIOGRAPHIES

0% G=2

G=4

G=5

G=8

G = 10

G = 20

G = waveband size

■ Figure 6. Cost reduction in hybrid hierarchical optical networks.

CONCLUSIONS In this article we outline the advantages of a hybrid hierarchical approach to optical networking. By leveraging the strengths of opto-electronic and optical technology along with the flexibility and cost efficiency of employing wavebands, this approach has the potential to become part of a scalable and profitable solution for long-haul and metro core networks. In a hierarchical optical network, the concept of uniform wavebands can be extended to nonuniform wavebands toward increasing the potential benefits. While promising high switching and routing performance in a cost-efficient manner, this concept comes with new challenges. On the node level, new architectural solutions have to be explored for further improvement of switching performance. On the network level, new elements (uniform and nonuniform wavebands) are introduced in traditional control and design problems. For hybrid hierarchical optical networks, the relationship between various components of the optical control plane becomes tighter and requires intelligent routing and resource (wavelength and waveband) assignment for optimal performance.

ACKNOWLEDGMENTS The authors thank the members of NEC Networking Research Labs and C&C Research Labs who contributed to this work. They are very grateful to Mr. K. Watanabe, general manager of NEC C&C Research Labs. for his guidance and encouragement. The authors also thank the anonymous reviewers for their helpful comments.

REFERENCES [1] R. Ramaswami and K. Sivarajan, Optical Networks: A Practical Perspective, Morgan Kaufmann, 1998. [2] L. Noirie, M. Vigoureux, and E. Dotaro, “Impact of Intermediate Traffic Grouping on the Dimensioning of Multi-Granularity Optical Networks,” Proc. OFC 2001.

94

[3] M. Lee et al., “Design of Hierarchical Crossconnect WDM Networks Employing a Two-Stage Multiplexing Scheme of Waveband and Wavelength,” IEEE JSAC, vol. 20, no. 1, 2002, pp. 166–71. [4] E. Ciaramella, “Introducing Wavelength Granularity to Reduce the Complexity of Optical Cross Connects,” IEEE Phot. Tech. Lett., vol. 12, no.6, June 2000, pp. 699–701. [5] K. Harada et al., “Hierarchical Optical Path Cross-Connect Systems for Large Scale WDM Networks,” Proc. 1999 OFC. [6] R. Lingampalli and P. Vengalam, “Effect of Wavelength and Waveband Grooming on an All-optical Networks with Single Layer Photonic Switching,” Proc OFC. 2002. [7] S. Araki et al., “Hybrid Cross-Connects and Hierarchical Optical Networks,” NEC R&D Journal, Apr. 2002. [8] Y. Suemura et al., “Waveband Routing in Optical Networks,” Proc. 2002 IEEE ICC. [9] R. Izmailov et al., “Hierarchical Optical Switching: A Node-Level Analysis,” Proc. 2002 Wksp. High Perf. Switching and Routing. [10] P. Ho and and H. Mouftah, “Path Selection with Tunnel Allocation in the Optical Internet Based on Generalized MPLS architecture,” Proc. IEEE GLOBECOM 2001.

RAUF IZMAILOV ([email protected]) received his M.S. degree in mathematics from Moscow State University, Russia, in 1984. He holds a Ph.D. degree in control science from the Institute of Control Science, Moscow, which he received in 1988, and where he worked since 1984 to 1989. From 1989 to 1991 he was with the Institute of Information Transmission Problems, Moscow, and from 1991 to 1993 he was with AT&T Bell Labs, Holmdel, NJ. In 1994 he joined NEC C&C Research Labs, Princeton, NJ, where he serves as department head of IP/Optical Networking. His research interests include routing and control problems in telecommunication networks, and optimization algorithms. SAMRAT GANGULY ([email protected]) received his B.Sc. degree in physics from the Indian Institute of Technology in 1994 and his M.E. in computer science from the Indian Institute of Science in 1998. He is pursuing a Ph.D. degree in computer science at Rutgers University. He joined NEC C&C Research Labs in 2000 and serves as a research engineer in the IP/Optical Networking department. His research interests are in the areas of routing and provisioning in IP/optical networks and QoS provisioning in wireless network. TING WANG ([email protected]) received his B.S. degree in electrical engineering from Beijing Polytecnic University, China, in 1982, his M.E. degree in electrical engineering from City University of New York in 1990, and a Ph.D. in electrical engineering from Nanjing University of Science and Technology in 1997. He joined NEC USA in 1991 and is now a senior research staff member with IP/Optical Networking at NEC C&C Research Labs, Princeton, New Jersey. His research interests include optical networking architectures, optical devices and subsystems, and high-speed opto-electronics in the applications of telecommunications. YOSHIHIKO SUEMURA ([email protected]) received his B.E. and M.E. degrees in electrical engineering from Kyoto University, Japan, in 1988 and 1990, respectively. In 1990 he joined NEC Corporation, Kawasaki, Japan, where he has been engaged in the research and development of optical interconnects, optical switching systems, and WDM optical networks. YOSHIHARU MAENO ([email protected]) received his B.S. and M.S. degrees in physics from Tokyo University, Japan, in 1991 and 1993, respectively. In 1993 he joined NEC Corporation, Kawasaki, Japan, where he has been engaged in the research and development of optical interconnections, WDM optical networks, and optical switching systems. SOICHIRO ARAKI ([email protected]) received his B.E. and M.E. degrees in electrical engineering from Kyoto University, Japan, in 1987 and 1989, respectively. In 1989 he joined NEC Corporation’s Opto-Electronics Research Laboratories, Kawasaki, Japan. In 1995, he spent a year as a visiting researcher at NEC Research Institute, Princeton, NJ, contributing to the analysis of communication performance in PC clusters. Since 1996 he has been developing a highspeed switching system based on optical switching techniques. He is now engaged in the research and development of WDM optical networks and WDM node architectures. He is a principal researcher in Networking Research Laboratories, NEC Corporation.

IEEE Communications Magazine • November 2002