OPtical networking is envisioned as the foremost solution. Waveband Switching for Dynamic Traffic Demands in Multi-granular Optical Networks

Transcription

1 1 Waveband Switching for Dynamic Traffic Demands in Multi-granular Optical Networks Xiaojun Cao, Vishal Anand, and Chunming Qiao Abstract Waveband Switching (WBS) in conjunction with Multi-Granular optical cross-connect (MG-OXC) architectures can reduce the cost and complexity of OXCs. In this paper, we study the performance of different MG-OXC architectures under dynamic traffic. In the case with online incremental traffic, we compare two MG-OXC architectures in terms of the blocking probability of new lightpath requests, and study the impact of port counts and traffic loads. We develop an on-line integer linear programming model (On-ILP), which minimizes the number of used ports and the request blocking probability, given a fixed number of wavelengths and MG-OXC architecture. The On- ILP optimizes the routing of new lightpaths so as to maximize lightpath grouping and reduce the port count given that existing traffic cannot be rearranged. We also propose a new efficient heuristic algorithm, called Maximum Overlap Ratio (MOR) to satisfy incremental traffic and compare it with the On-ILP, First- Fit, and Random-Fit algorithms. Our results and analysis indicate that using WBS with MG-OXCs can reduce the size (and hence the cost) of switching fabrics compared to using ordinary OXCs. Based on the results and observations in the incremental traffic case, we further study the performance of a particular MG- OXC architecture under fully dynamic or fluctuating traffic. Our simulations show that the proposed heuristic algorithm WAPG which groups wavelengths to bands and uses wavelength converters efficiently under fluctuating traffic, significantly outperforms other heuristic algorithms. I. INTRODUCTION OPtical networking is envisioned as the foremost solution to meet the growing data traffic from the ever-increasing Internet applications/services such as VoIP, IPTV, Internet games, and large-scale science collaborations. While the use of Wavelength-division multiplexing (WDM) technology has significantly increased the available bandwidth in backbone networks, the rapid advances in Dense WDM technology with hundreds of wavelengths per fiber and world-wide fiber deployment have brought about a tremendous increase in the cost and size of electronic cross-connects or DXCs (e.g. OEO grooming switches). Optical (photonic) cross-connects (OXCs) that switch bypass traffic all-optically are useful in reducing the cost and size of the OEO grooming switches. However, Partial and preliminary results appeared in the Proceedings of IEEE INFOCOM, Hongkong, March X. Cao is with the Department of NSSA, Rochester Institute of Technology, Rochester, NY 14623, USA, ( cao@mail.rit.edu). V. Anand is with the Department of CS, SUNY College at Brockport, Brockport, NY USA ( vanand@brockport.edu). C. Qiao is with the Department of CSE, State University of New York at Buffalo, Buffalo, NY USA ( qiao@computer.org). X. Cao is supported in part by NSF CNS , V. Anand is supported in part by the Scholarly Incentive Award at SUNY Brockport and C. Qiao is supported in part by NSF when the number of wavelengths is large traditional OXCs that switch traffic only at the wavelength granularity themselves can become huge (i.e. requiring a large number of wavelength ports), resulting in increased cost and control complexity. Waveband Switching (WBS) in conjunction with new Multi- Granular optical cross-connects (or MG-OXCs) that can switch traffic at fiber, waveband and wavelength granularities [1], [2], [3], [4], [5], [6], [7], [8], has been proposed to reduce this cost and complexity. The main idea of WBS is to group several wavelengths together as a band, and switch the band using a single port whenever possible (e.g. as long as it carries only bypass or express traffic), and demultiplex it to switch the individual wavelengths only when some traffic needs to be added/dropped. As the bypass traffic accounts for upto 60% to 80% of the total traffic in the backbone, only a limited number of fibers and bands need to be demultiplexed into wavelengths. Thus, not only the size of wavelength cross-connects, but also the overall number of ports and complexity of the MG-OXCs can be reduced by using waveband switching. The concept of WBS based on two stage multiplexing was applied to WDM ring networks in [5], while its merits such as small-scale modularity, cross-talk and complexity reduction were summarized in [4]. A Three-Layer switching fabric consisting of a fiber cross-connect (FXC), a band cross-connect (BXC) and a wavelength cross-connect (WXC) was presented in [2], and the application of such Three- Layer MG-OXC architectures to metro-area networks was described in [9]. For such Multi-Layer MG-OXCs, limited analytic work for some specific traffic patterns in rings was done in [10], [11], [12] while [13] studied the benefit of non-uniform waveband hierarchy. Hybrid hierarchical switches with all-optical waveband switching and OEO traffic grooming were shown to reduce cost in [14], [15], [16] while a quantitative investigation on wavelength grouping granularity is presented in [17]. The authors of [1], [18] presented MILPbased approaches for the design of a two-layer MG-OXC network using a simple lightpath grouping strategy, which does not take full advantage of the benefits of wavebanding. In our prior research [3], the most powerful lightpath grouping strategy, i.e. sub-path grouping for a Three-Layer MG-OXC architecture was adopted. In particular, we provided a general Integer Linear Programming (ILP) model, and an efficient heuristic called Balanced Path routing with Heavy-Traffic first waveband assignment (BPHT) for off-line static traffic in [3]. The ILP model and the BPHT heuristic were also extended to multi-fiber systems in [19]. Recently, the authors in [7] proposed a Single-Layer MG-OXC architecture for WBS.

2 2 However, no detailed algorithms or comparisons with other architectures was given. In [14], we discussed the differences between WBS and traditional wavelength routed networks (WRN) and provided an overview of the issues related to WBS such as survivability and wavelength/waveband conversion. This work provided only a qualitative comparison of the Single-Layer and Multi-Layer MG-OXC architectures, and for the first time gave numerical results for the on-line case, but without any detailed algorithms. Issues related to multigranularity optical switching and waveband grouping under the Generalized Multi-Protocol Label Switching (GMPLS) framework, such as signaling protocols and Link Management Protocols, have been partially addressed in [20], [21]. While a significant number of research focused on wavelength routed networks (see for example, [22], [23], [24]), most existing work on WBS has been limited to only a specific MG-OXC architecture for the off-line case (where a set of lightpath requests is known a priori). In this work, we quantitatively compare the performance of the Single-Layer and the Three-Layer MG-OXCs 1. In the case with incremental traffic, new lightpath requests need to be processed one at a time without knowledge of any future requests while existing connections stay indefinitely and are non-rearrangeable. We develop an on-line ILP model (On-ILP), which minimizes the used ports and the request blocking probability, given a fixed number of wavelengths and MG-OXC size. As our analysis and simulation indicate that multi-layer MG-OXC is more suitable for dynamic traffic, we further study the performance of multi-layer MG-OXC architecture under fully dynamic traffic (i.e. connections are established and released randomly). Although several studies (see for example, [25], [26], [27], [28]) have shown that wavelength conversion offers significant benefit in WRNs by reducing the blocking probability of Wavelength Routed Networks (WRNs) with dynamic traffic, the effect of wavelength conversion in WBS networks has not been addressed. Hence, we also propose an efficient heuristic algorithm called Waveband Assignment with Path-Graph (WAPG) that efficiently uses wavelength converter resources and at the same time maximizes the benefit of wavebanding, to carry dynamic traffic and compare their performance with other heuristics such as First- Fit and Random-Fit algorithms. Our results show that the proposed heuristic algorithm significantly outperforms other heuristic algorithms. Our study provides valuable insights into the trade-offs between port counts and blocking probability. This performance evaluation and our proposed heuristics are especially useful for minimizing the number of used ports in a WDM network, and thus the network operating costs, while achieving a low blocking probability of requests. To the best of our knowledge, this is the first study of WBS algorithms (ILP and heuristics) for on-line traffic, and also the first quantitative comparison of the Three-Layer and Single-Layer MG-OXC architectures. This paper is organized as follows. Section II describes 1 From now on, we use the terms Multi-Layer MG-OXC and Three-Layer MG-OXC interchangeably. and compares the Three-Layer and Single-Layer MG-OXC architectures for WBS. In Section III we present our Online (for incremental traffic) ILP and heuristic algorithm for both the Three and Single-Layer MG-OXC, and compare their performance using simulation. In Section IV, we study the performance of reconfigurable Three-Layer MG-OXC for fully dynamic traffic. Finally, we conclude the paper in Section V, with a summary of its major contributions. II. MULTI-GRANULAR OPTICAL CROSS-CONNECT ARCHITECTURES Thanks to the development of OXCs, optical WDM networks are evolving from interconnected SONET ring to arbitrary-mesh topologies. In traditional optical networks, wavelengths terminate at, or transparently pass-through a node using an ordinary-oxc. Such ordinary-oxcs switch each individual wavelength using one port. On the other hand, in WBS networks several wavelengths are grouped together as a band, and switched as a single entity (i.e. using a single port) whenever possible. A band is demultiplexed into individual wavelengths if and only if necessary, e.g. when the band carries at least one lightpath that needs to be dropped or added. A complementary hardware is MG-OXC that not only can switch traffic at multiple levels (or granularities) such as fiber, waveband, and individual wavelength, but also may add/drop traffic at multiple levels, as well as multiplex/demultiplex traffic from one level to another. F add B drop Fig. 1. : DGG βy %7: :;& %;& );& : GURS :7% %7) )7% ; < αx :;& /D\HU B add %;& /D\HU );& /D\HU F drop Three-Layer Multi-Granular Photonic Cross-connect The MG-OXC is a key element for routing high speed WDM data traffic in a multi-granular optical network. While reducing its size has been a major concern, it is also important to devise node architectures that are flexible (reconfigurable) yet cost-effective. Two principle MG-OXC architectures: the Three-Layer and Single-Layer, have been proposed in literature.

3 3 ) DGG % DGG : DGG : GURS % GURS ) GURS Q αx βy Fig. 2. :;& %;& );& Single-Layer Multi-Granular Photonic Cross-connect A. Three-Layer MG-OXC Figure 1 shows a typical Three-Layer MG-OXC, which includes the FXC, BXC and WXC layers. As shown in the figure, the WXC, BXC layers consist of cross-connect(s) and multiplexer(s)/demultiplexer(s). The WXC layer includes a wavelength cross-connect (WXC) that is used to switch bypass lightpaths. To add/drop wavelengths from the WXC layer, we need W add /W drop ports and multiplexers/demultiplexers. At the BXC layer, the waveband cross-connect (BXC), B add and B drop ports are used for bypass wavebands, added wavebands and dropped wavebands respectively. In addition, BTW ports are used to demultiplex wavebands to WXC layer and WTB ports are used to multiplex wavelengths from WXC layer to bands. Similarly, fiber cross-connect (FXC)/F add /F drop ports are used to switch/add/drop fibers at the FXC layer. FTB and BTF ports are used to demultiplex fibers to wavebands, and multiplex wavebands to fibers, respectively. In order to reduce the number of ports, the MG-OXC switches a fiber using one port (space switching) at the FXC cross-connect if none of its wavelengths is used to add or drop a lightpath. Otherwise, it will demultiplex the fiber into bands, and switch an entire band using one port at the BXC cross-connect if none of its wavelengths needs to be added or dropped. In other words, only the band(s) whose wavelengths need to be added or dropped will be demultiplexed, and only the wavelengths in those bands that carry bypass traffic need to be switched using the WXC. This is in contrast to the ordinary-oxcs, which needs to switch every wavelength individually using one port. This multi-layer architecture allows dynamic selection of fibers for multiplexing/demultiplexing from FXC layer to the BXC layer, and bands for multiplexing/demultiplexing from BXC to the WXC layer. For example, at the FXC layer, as long as there is a free FTB port, any fiber can be demultiplexed into bands. Similarly, at the BXC layer any band can be demultiplexed to wavelengths using a free BTW port by appropriately configuring the FXC, BXC cross-connects and associated demultiplexers. Unlike in the off-line case [19], where the MG-OXC can Q have as many port as needed to guarantee that all the demands are satisfied, here, the MG-OXC has only a predetermined limited port count to accommodate dynamic traffic. More specifically, let X denote the number of incoming fibers, Y the number of BXC ports from FTB demultiplexers, α 1 the ratio of fibers (to the total number of fibers) that can be demultiplexed into bands using FTB ports, and similarly, β 1 the ratio of bands that can be demultiplexed to wavelengths using BTW ports. Such Three-Layer MG-OXC architecture is reconfigurable (and hence flexible) in that any αx fibers can be demultiplexed into bands and any βy of these bands can be demultiplexed into wavelengths simultaneously by appropriately configuring the MG-OXC. We show that even with limited reconfiguration (i.e. α<1 and β<1), we can use an intelligent algorithm (e.g. the proposed MOR, WAPG for routing and wavelength/waveband assignment) to considerably reduce the port count required to satisfy dynamic traffic with an acceptable request blocking probability. The total number of ports at such a reconfigurable, Three- Layer MG-OXC node n can be calculated as in Equation (1). MG OXC n = (1+α) X +(1+β) Y +β Y W +W add/drop (1) Note that when α =1,β =1, there is no limitation on the number of fibers/bands that can be multiplexed/demultiplexed, and hence, the blocking of a lightpath request can only come from the limited number of wavelengths as in an ordinary- OXC network. If we consider single-fiber systems and let δ be the degree of node n, wehavex = δ, Y = α X B. For an ordinary-oxc that only switches individual wavelengths, the number of ports at node n is OXC n = δ B W + W add/drop. Accordingly, if we ignore the W add/drop ports (which are common to both the Three-Layer reconfigurable MG-OXC and ordinary-oxc), Equation (2) gives the ratio of the port count in a Three-Layer MG-OXC to the port count in an ordinary-oxc, denoted by T 3. T 3 = (1 + α) δ +(1+β) α δ B + α β δ B W (1 + β) α β α + W δ B W B. Single-Layer MG-OXC Compared to the previously described Three-Layer MG- OXC, the one shown in Figure 2 is a Single-Layer MG- OXC that has only one common switching fabric[7]. This switching matrix includes three logical divisions corresponding to the FXC, BXC and WXC, respectively. However, the major differences (from the Three-Layer MG-OXC) are the elimination of FTB/BTW demultiplexers and BTF/WTB multiplexers between the different layers, which results in a simpler architecture to implement, configure and control. Another advantage of this Single-Layer MG-OXC is better signal quality because all lightpaths go through only one switching fabric, whereas in the Three-Layer MG-OXC, some of them may go through as many as three switching fabrics (i.e. FXC, BXC and WXC). As a trade-off, some incoming fibers, e.g. fiber n (see Figure 2), are pre-configured as designated fibers. Only these designated fiber(s) can have some of its bands (2)

4 4 dropped while the remaining bands bypass the node, all other non-designated incoming fibers (e.g. fibers 1 and 2) have to have all the bands either bypass the node entirely or be dropped entirely. Similarly, within these designated fiber(s), only designated band(s) can have some of its wavelengths dropped while the remaining bands bypass the node. Thus, the Single-Layer MG-OXC is simple, but not flexible in that it does not allow lightpaths to be multiplexed/demultiplexed and grouped into bands arbitrarily, which may result in inefficient utilization of network resources. More specifically, in WBS networks with Single-Layer MG-OXCs, an appropriate WBS algorithm needs to make sure that the lightpaths to be dropped at a Single-Layer MG-OXC will be assigned wavelengths that belong to a designated fiber/band. Clearly, this may not be always possible if there is only a limited number of designated fibers/bands, especially in the case of on-line traffic where global optimization for all lightpath demands is often difficult (if not impossible) to achieve. For this reason, a network with Three-Layer MG-OXCs may in fact require fewer ports and wavelengths in order to satisfy all the on-line lightpath demands, or result in a better blocking performance (i.e. a lower blocking probability) for a given set of on-line lightpath demands with the same number of wavelength and ports. Similar to Equation (2), if we limit only αx fibers can be demultiplexed into bands and βy of these bands can be demultiplexed into wavelengths simultaneously, Equation (3) gives the ratio of the port count in a Single-Layer MG-OXC to the port count in an ordinary-oxc, denoted by T 1. The difference between T 3 and T 1 is due to the fact that there are no FTB/BTF and BTW/WTB ports in the Single-Layer MG-OXC architecture, which are present in the Three-Layer MG-OXC architecture. (1 β) α T 1 β α + (3) W From Equations (2) and (3), we see that, in order to reduce the port count by using MG-OXCs instead of ordinary-oxcs, the values of α and β need to be constrained so as to ensure that T 3 < 1 and T 1 < 1. For single-fiber systems, it is necessary to set α =1to allow any fiber to be demultiplexed to bands (otherwise, the blocking probability is too high). However, we can/should limit the value of β to be less than 1 by allowing only a limited number of bands (i.e. βy ) to be demultiplexed into wavelengths simultaneously. Note that, as alluded to above, an MG-OXC (Three-Layer or Single-Layer) can reduce the port count or achieve a low blocking probability only when efficient WBS algorithms are employed. In the remainder of the paper, we will develop ILP-based mathematical models and heuristic algorithms for the Three-Layer and Single-Layer MG-OXC networks, and compare the two architectures quantitatively. Hereafter, we concentrate on one of the proposed WBS schemes in [3], wherein each fiber has a fixed number (B) bands and each band has a fixed number (W) as well as a fixed set of wavelengths. Note that the ILP model and heuristic algorithms developed in this paper can be extended to the other WBS schemes (e.g. allowing variable number of bands per fiber) as well. III. Waveband Switching with Incremental Traffic The off-line problem is meaningful when building a greenfield WBS network. Another challenging problem is how to design WBS algorithms for the on-line case, given an existing MG-OXC architecture and network. In this section, we develop an ILP model to accommodate new lightpaths assuming that existing connections stay indefinitely and are non-rearrangeable. We also assume that there is no wavelength conversion in our models for Incremental traffic. A. On-Line ILP model (On-ILP) To reduce the computation complexity to a reasonable level, we develop our ILP model based on K-shortest path algorithms to limit the number of possible routing. Below, we first present the near-optimal ILP model for the Three- Layer MG-OXC and then adapt it to suite the Single-Layer MG-OXC. Notations: The following parameters are used by the ILP model. I n : Set of input fibers at node n (excluding those for local add); O n : Set of output fibers at node n (excluding those for local drop); A n : Set of local add fibers at node n, including those used at the WXC, BXC and FXC layer; D n : Set of local drop fibers at node n, including those used at the WXC, BXC and FXC layer; IA n : I n An. This set includes all incoming fibers (local and non-local) at node n; OD n : O n Dn. This set includes all outgoing fibers (local and non-local) at node n; Λ b : Set of wavelengths in band b; X: Number of wavelengths per fiber; B: Number of wavelength bands per fiber; W: Number of wavelengths per wavelength band (X = B W); P: Set of node pairs having non-zero traffic demand. T[p]: Traffic matrix whose element t p is an integer, representing the traffic demand (i.e. number of lightpaths) for the node pair p; K: Maximum number of paths that can be used for routing for a node pair; L k,p : The set of links along the k th shortest path of the EW n,w i,o EB n,b i,o : node pair p; : =1, denotes there exists a lightpath using wavelength w on an incoming fiber i through the WXC layer onto an outgoing fiber o at node n; =1, denotes there exist lightpaths using waveband b (b [1, 2,...,B]) on an incoming fiber i through the BXC layer onto an outgoing fiber o at node n;

5 5 EF n i,o : =1, denotes there exist lightpaths using an incoming fiber i through the FXC layer onto an outgoing fiber o at node n. ILP Variables: To facilitate the presentation and understanding of our ILP model, we first define variable Vk,p w to help select one of the several shortest-paths. V w k,p : 1 if a lightpath for the node pair p uses wavelength w along the k th shortest path, and 0 otherwise; To describe the drop/bypass/add traffic (lightpath) at a node, the following four variables: S n,w i,o, W n,w i,o, Bn,b i,o and Fi,o n are used, where I n An is called incoming fiber and O n Dn is called outgoing fiber. More specifically, when i I n,o O n, these variables represent bypass traffic; add traffic when i A n,o O n, and drop traffic when i I n,o D n (note that the case when i A n,o O n does not make sense). S n,w i,o : W n,w i,o B n,b i,o : F n i,o : 1 if at node n, there is a lightpath using wavelength w on an incoming fiber i to outgoing fiber o, and 0 otherwise; : 1 if node n has a lightpath using wavelength w on an incoming fiber i through the WXC layer onto an outgoing fiber o, and 0 otherwise; 1 if node n has a set of lightpaths using waveband b (b [1, 2,...,B]) on an incoming fiber i through the BXC layer onto an outgoing fiber o, and 0 otherwise; 1 if node n has a set of lightpaths using an incoming fiber i through the FXC layer onto an outgoing fiber o, and 0 otherwise; The following four additional variables are also defined for describing the multiplexing/demultiplexing at the FXC, BXC and WXC layers. FTBi n: 1 if input fiber i (i I n) needs to be demultiplexed into bands at node n, and 0 otherwise; BTW n,b i : 1 if band b on input fiber i (i I n ) needs to be demultiplexed into wavelengths at node n, and 0 otherwise; BTFo n : 1 if a band needs to be multiplexed onto an output fiber o (o O n ) at node n, and 0 otherwise; WTBo n,b : 1 if a wavelength needs to be multiplexed on to band b of an output fiber o (o O n ) at node n, and 0 otherwise; Objective Function: There are two reasonable objectives. The first is to minimize the total used MG-OXC ports in the network, which is: min [ n,i,o,w W n,w i,o + + n,i,o n,i,o,b (B n,b i,o + WTBn,b o + BTW n,b i ) (Fi,o n + BTF o n + FTBi n )] (4) The second objective is to minimize the maximum port count (node size) over all the nodes in the network. This can be formulated as: min max [ W n,w n i,o + (B n,b i,o + WTBn,b o + BTW n,b i ) i,o,w i,o,b + (Fi,o n + BTF o n + FTBi n )] (5) i,o Constraints: The following Equations (6) - (8) ensure that the coming traffic is satisfied, and each is assigned wavelength resources along its route while existing connections remain unaffected. W n,w i,o S n,w i,o w,k V w k,p = t p p; (6) V w k,p k, p, w, i, o L k,p ; (7) EW n,w i,o, Bn,b i,o EBn,b i,o, Fn i,o EFi,o n i, o, n; (8) For Waveband Switching, we need the following additional constraints. 1 Fi,o+B n n,b n,w i,o +Wi,o S n,w i,o w Λ b,i IA n,o OD n ; (9) 1 F n i,o + o 1 o S n,w i,o 1, 1 B n,b i,o + S n,w i,o 1, o 1 o 1 F n i,o + i 1 i S n,w i 1,o w, i, o; (10) 1 B n,b i,o + S n,w i i, o, w Λ 1,o b; i 1 i (11) Constraints (9) (11) ensure that if a lightpath uses wavelength w belonging to band b of incoming fiber i and outgoing fiber o (i.e. S n,w i,o =1), then at node n, exactly one of FXC, BXC and WXC cross-connect port will be used for switching this lightpath when it is a bypass (i.e. i I n,o O n )or exactly one of F add, B add and W add port will be used for adding this lightpath when it is added (i.e. i A n,o O n )or exactly one of F drop, B drop and W drop port will be used for dropping this lightpath when it is dropped (i.e. i I n,o D n ) In addition, the constraint below ensures that a wavelength w at node n switched or added at the WXC layer has to pass a WTB multiplexer to the BXC layer. At the same time, every

6 6 band from a WTB multiplexer has to pass a BTF multiplexer before it can leave node n. BTF n o WTB n,b o W n,w i,o w Λ b,o O n,i IA n ; (12) Similarly, Equation (13) below specifies that a wavelength w switched or dropped at the WXC layer has to come from BXC layer using a BTW demultiplexer, and in addition every band demultiplexed by BTW can only come from a FTB demultiplexer. FTB n i BTW n,b i W n,w i,o w Λ b,o OD n,i I n ; (13) Any bypass or add bands should pass a BTF multiplexer as specified in equation (14) and similarly, any drop or bypass band can only come from a FTB demultiplexer as specified in Equation (15). BTF n o B n,b i,o o O n,i IA n ; (14) FTBi n B n,b i,o o OD n,i I n ; (15) Finally, Equations (16) and (17) are the constraints for Three-Layer MG-OXC architecture, which constrain the total number of bands that can be demultiplexed/multiplexed using the BTW/WTB ports 2. o,b i,b WTB n,b o δ β n (16) BTW n,b i δ β n (17) The above ILP model is used to minimize additional number of ports and multiplexers/demultiplexers in a network with Three-Layer MG-OXCs. By ignoring (i.e. not counting) the ports for FTB/BTF and BTW/WTB, we can also apply it to WBS networks with Single-Layer MG-OXCs. However, in the Single-Layer MG-OXC architecture, the choice of the designated bands (i.e. which bands can be demultiplexed into wavelengths) is critical. Since the traffic carried by a designated band at one node may bypass at another node, we cannot set the same band(s) as the designated bands for every node in the network. Particularly, when the traffic is dynamic (i.e., random) it is almost impossible to predict how future lightpath(s) will be routed and grouped into which band(s), which may eventually need to be added/dropped or multiplexed/demultiplexed at different nodes. Ideally we would want this band(s) to be designated. However, due to the fact that the traffic is essentially random, it is not possible to determine which of these bands should be designated. Also note that to save on ports we want to make the number of designated bands small. Therefore, the subset of designated bands at each node will be randomly selected in this study. We use the following two additional parameters denoting the setting of designated band(s) and associated multiplexers/demultiplexers at every node. 2 Since we have set α =1, there is no constraint on the number of FTB/BTF ports. EBTW n,b i : 1 if band b on input fiber i (i I n ) can be demultiplexed into wavelengths at node n, and 0 otherwise; EWTBo n,b : 1 if wavelengths can be multiplexed on to band b of an output fiber o (o O n ) at node n, and 0 otherwise; For a given β (or T 1 ) value, the probability that the above variables will be set to 1 (i.e. is a designated band) for a given band b is equal to β. Similarly, Equations (18) and (19) constrain the number of designated bands at a node in the Single-Layer MG-OXC. WTBo n,b BTW n,b i EWTB n,b o o, n, b (18) EBTW n,b i i, n, b (19) For the above K-shortest path based On-ILP model, if we set K =1, the ILP will use only one (i.e. the first shortest path) for routing the lightpath between every node pair, and assign wavelengths such that the used port count is minimized. However, if we set K =, the ILP will search all possible routes. By restricting K, we reduce the search space of the ILP, so that we can get a near-optimal solution in a reasonable time. Our experiments show that the On-ILP model can be applied to moderate size MG-OXC networks. The On-ILP model can be extended and applied to WBS networks with static traffic. We note that in the case of on-line traffic the On-ILP can be extended to obtain global optimization i.e., for each new traffic demand compute the paths and wavelength grouping to minimize the total port counts by setting K = and allow rearrangement of all existing connections. However, our experiments show that the ILP for global optimization is too time consuming for even moderate network size and demands. Thus one has to seek near-optimal solutions such as the On- ILP model. From our work in [29]we notice that Single-Layer MG-OXCs are more beneficial than Three-Layer MG-OXCs in terms of reducing node size in WBS networks with static traffic. B. On-line Heuristic Algorithm for Incremental Traffic We now illustrate our Maximum Overlap Ratio (MOR) algorithm, whose objective is to minimize the request blocking probability by performing efficient routing and wavelength (and waveband) assignment in WBS networks for on-line traffic. Given that there is no wavelength or waveband conversion in the MG-OXCs, we model a WBS network (for example Figure 3(a)) using B band-graphs (one for each band) as in Figure 3(b). The nodes in each band-graph correspond to the nodes in the physical network topology, while the links between the nodes correspond to the existence of that band between the nodes. For a new lightpath demand for a node pair p, we first find up to K-shortest paths (R1,R b 2, b...rk b ) in each band-graph b, such that each path Rk b has at least one free wavelength that can be used to establish the lightpath. We then determine the

7 7 weight denoted by Q b k of the kth (1 k K) shortest path in band b. Fig. 3. graphs D λ 0 λ 2 λ 4 H[LVWLQJOLJKWSDWKV E E E E 6 λ 5 QHZGHPDQG Determining the weight of each of the K-shortest paths using band Intuitively, in order to satisfy a new lightpath demand with as few additional ports as possible, it is better to route a new lightpath along a path Rk b that has maximum number of links in common with all the existing lightpaths, established along the same path in band b. On the other hand, to avoid the wastage of wavelength resources, it is better to route along the shortest possible path. To achieve a balance, we set Q b k to be L/H, where H is the number of hops (in path Rk b ) and L is the sum of overlap length (number of links in common with all existing lightpaths) in band b. Algorithm MOR chooses a path Rk b that has the maximum weight Qb k, to route the new lightpath and assigns the first available wavelength along Rk b to the lightpath. For example, suppose each fiber has three bands b 0, b 1 and b 2, and each band has two wavelengths:(λ 0,λ 1 ) b 0, (λ 2,λ 3 ) b 1,(λ 4,λ 5 ) b 2 as shown in Figure 3(b). Now assume a new request for a lightpath from node 0 to node 7 arrives, for which three paths R bi 1, Rbi 2 and Rbi 3 are available in each band b i, where i = {1, 2, 3}, as shown in Figure 3(a). The maximum weight is Q b2 k 2 =3/4 (note that Q b0 k 1 =1/4 and Q b1 k 3 =3/5), hence the new lightpath will use λ 5 in band b 2 on path , as indicated by the dashed line in Figure 3(b). Such grouping of lightpaths allows the use of the already existing (in use) ports at node S 4,S 5 and S 6, leaving more unused ports for future requests. One of the variations of MOR is to maximize L (the overlap length) only, by setting Q k b equal to L instead of L/H. We have compared MOR and its variations and found that maximizing the L/H ratio performs the best, and hence show only the results obtained from MOR below. C. Numerical Results In this section, we study the performance of various on-line WBS algorithms, and compare the performance of the Three-Layer and Single-Layer MG-OXC networks having the same topology as the 14-node NSF network. Figures 4 to 7 illustrate how the request blocking probability and the used wavelength resources vary with changing β (i.e., the ratio of bands that can be demultiplexed/multiplexed). These results are obtained when the number of wavelengths per band is fixed at W =8and the number of bands per fiber is B =10. Similar results for different values of W and B are omitted due to space limitation. In addition, the amount of wavelength resources is expressed in terms of wavelength hops. For example, if a traffic request requires two lightpaths to be setup between the source and destination that are 3 links apart, then the number of wavelength hops is 2 3=6. Accordingly, the wavelength-hop ratio is defined as W = W avelength hops required for W BS using MG OXCs Minimal wavelength hops required in W RNs using ordinary OXCs. We set K =3for MOR, On-ILP and heuristics Random-Fit and First-Fit. The heuristic Random-Fit routes the new lightpath request along the shortest possible path assigning it a random wavelength (in a random band). Heuristic First-Fit, on the other hand, routes the new lightpath request along the shortest path assigning it the first available wavelength. Fig. 4. Blocking Probability in Three-Layer MG-OXCs - Med load 1) Three-Layer MG-OXC: From Figures 4 and 5, we note that when β 0.45 (i.e. T ), MOR achieves the lowest blocking probability. Increasing β to greater than 0.45 does not help in reducing the blocking probability any further because now blocking occurs only due to limited wavelength resources and not due to limited reconfiguration flexibility (e.g. ports). In other words, no more than 45% of the bands need to be demultiplexed into wavelengths (and increasing β further will only unnecessarily increase the port count). In fact we need to keep β<0.8 in order to take advantage of Three-Layer MG- OXCs (i.e. ensure T 3 1). One of the practical implications of this result is that one may want to build-in about β = 45% (but not more) BTW ports, and use them (e.g. to accommodate incremental traffic) when needed. When β = 45%, the ratio of the port count in a Three-Layer MG-OXC to the port count in an ordinary-oxc is T , which indicates we can achieve a 35% savings in the number of ports when using Three-Layer MG-OXCs instead of ordinary-oxcs.

8 8 Fig. 5. Blocking Probability in Three-Layer MG-OXCs - High load We can also see that when the traffic load is moderate, MOR can achieve the lowest blocking probability with First- Fit being almost as good. Under heavy load, MOR performs the best as long as the port number ratio T 3 is less than 0.75, above which First-Fit performs better (i.e. has a lower blocking probability). This is mainly due to the fact that in the process of reducing the number of used ports, MOR chooses longer paths which consume more wavelength resources (i.e., larger wavelength-hop ratio W ) than First-Fit as shown in Figure 6. Hence, when T 3 is large and the load is high, MOR experiences more blocking (compared to First-Fit) due to the lack of wavelength resources. Furthermore, we can see that MOR is better than On-ILP, and much better than Random-Fit in terms of reducing the request blocking probability for a given value of T 3 in WBS networks. The reason for the poor performance of Random-Fit is that, unlike First-Fit it does not take waveband grouping into consideration. First-Fit, on the other hand, is very likely to assign wavelengths to lightpaths sequentially, which helps in wavebanding and thus reducing the number of used ports. As for On-ILP, since its objective is to minimize the number of additional ports for each new request, it cannot minimize the overall port counts by performing a global optimization over all lightpath requests. In other words, the On-ILP is short-sighted in that it will assign paths and wavelengths to the initial set of traffic demands so as to minimize the port count in the beginning. However, this initial greedy path and wavelength assignment, hurts its performance, when it has to assign more ports to future traffic demands, hence increasing its blocking probability. Fig. 6. Fig. 7. Used Wavelength-hop ratio in Three-Layer MG-OXCs Used Wavelength-hop ratio in Single-Layer MG-OXCs 2) Single-Layer MG-OXC: The fact that in Single-Layer MG-OXCs the designated bands are allocated randomly at different nodes reduces the chance of wavebanding and hence increases the blocking probability considerably. For example, it may happen that a node has no additional designated bands available to add/drop traffic even though there are enough resources (e.g. ports or wavelengths) at the intermediate nodes/links. Hence, the blocking probability of the Single- Layer MG-OXC network is higher than that of the Three-Layer MG-OXC network. In fact, even at a low load there is still a non-negligible blocking probability in the Single-Layer MG-OXC network when T (as shown in Figure 8), while the blocking probability of the Three-Layer MG-OXC network is close to zero (except when using the Random-Fit algorithm) and hence not shown. When the load is high, the blocking probability of the Single-Layer MG-OXC network is close to 1 (not shown). Figure 9 shows the request blocking probability of Single- Layer MG-OXC under a medium load. Compared to Figure 4, we notice that the blocking probability of the Three-Layer MG-OXC network is much lower (up to 10 times) than that of the Single-Layer MG-OXC network when both are equipped with the same number of ports. For example, when the ratio T 3 = T 1 =0.7, the blocking probability of the Three-Layer MG-OXC network is less than 0.02 (when using MOR) while

9 9 Fig. 8. Blocking Probability in Single-Layer MG-OXCs - Low load Wavelength conversion has been shown to be useful in wavelength routed optical networks, however, no work addresses the use of wavelength conversion in WBS networks. Since the technology for all-optical wavelength conversion is more readily available than waveband conversion, which is still an immature technology. We assume that wavelength conversion can only occur at the wavelength cross-connect (WXC) layer within a Three-Layer MG-OXC as shown in Figure 1. Wavelength conversion is particular interesting in WBS networks, as it can ease wavelength requirement and facilitate waveband assignment but may also cause some negative effects. There are two major unique issues related to wavelength conversion in WBS networks. Firstly, request blocking may come from not only the limited number of wavelengths (and wavelength conversion capability), but also the limited number of ports at the MG-OXCs. Secondly, even though a wavelength converter may be available for use to satisfy a request, performing wavelength conversion at WXC layer requires all the wavelengths in a band to be demultiplexed and hence consuming more ports, which in turn, may cause more blocking in networks with limited reconfigurable MG-OXCs (i.e. α<1 and β < 1). Therefore, how to effectively use the wavelength converters while satisfying fluctuating traffic is critical in WBS networks. A. Waveband Assignment with Path-Graph (WAPG) Fig. 9. Blocking Probability in Three-Layer MG-OXCs - Med load that of the Single-Layer MG-OXC network is about 0.2, which indicates that Three-Layer MG-OXCs are more suitable for dynamic traffic. As in the case of using Three-Layer MG-OXCs (see Figure 6), there is also a trade-off between the used ports and WHs when using Single-Layer MG-OXCs as shown in Figure 7. IV. WAVEBAND SWITCHING WITH FULLY DYNAMIC TRAFFIC Since results in Section III-C show that with dynamic incremental traffic the Three-Layer MG-OXCs outperforms the Single-Layer MG-OXCs by a large margin, hereafter, we concentrate only on Three-Layer MG-OXC architecture. In this section, we propose an efficient algorithm for Three-Layer MG-OXC in WBS networks with wavelength conversion under fully dynamic traffic conditions, and evaluate its performance. In this section, we propose a heuristic, namely Waveband Assignment with Path-Graph, or WAPG, that reduces the blocking probability by efficiently routing and assigning wavelengths to new lightpath requests and grouping them into bands such that the number of used wavelength converters is reduced. For illustration purpose, we assume that fixed routing (shortest path first or constrained shortest path first) is used. We note that our algorithm can also be applied to different routing schemes. For lightpath requests equal to or larger than the size of a band, WAPG tries to allocate one or more band paths for the requests, and then allocates individual wavelengths for the remaining requests as shown in Algorithm WAPG. For a lightpath request using path l, s 0 s 1,... s i s i+1... s n, H is the number of hops along the path and each link has X wavelengths, partitioned into B bands, each consisting of W wavelengths. Let 1 d X be the number of wavelength converters per link. Let Λ={w 1,w 2,..., w λ,..., w X } be the set of wavelengths, and b the index of waveband set = {1, 2,..., X/B} on each link. Then, wavelength 1 λ X belongs to band b = λ/b. We model a path l using X layers of path-graph (one for each wavelength). The nodes in each layer of the path-graph correspond to the nodes in the network topology. For a given path-graph (λ), the links between the nodes in the same layer correspond to the existence of that wavelength between the physical nodes while the links between different path-graph layers imply the existence of wavelength converters at the physical nodes.

10 10 Algorithm WAPG for each node s i on the path l, each wavelength w λ do Create logical node s λ i in the path-graph. if there are wavelength converters from λ to λ at the node then Create logical link between node s λ i and s λ i in the path-graph. Set the weight of the logical link s λ i s λ i as w(s λ i,sλ i )=X H. end if end for for each link s i s i+1 on the path l, each wavelength w λ do if wavelength w λ is available on link s i s i+1 then Create logical link between node s λ i and s λ i+1 in the path-graph. Set the weight of the logical link s λ i s λ i+1 as w(s λ i,sλ i+1 )=λ. end if end for Create logical node s, d and links s s λ 0, s λ n d. Set the weight of the links s s λ 0 and s λ n d as 0. Use Dijkstra algorithm search proper wavelengths and check if the constraints on the wavelength conversion are met to accommodate the new request. Based on previous observation that FirstFit wavelength assignment facilitates grouping wavelengths into bands and hence helps reducing the blocking probability in the case with incremental traffic, we construct the path-graphs and set the weight of the links in each wavelength layer to be the index number of the wavelength (as shown in algorithm WAPG).We then apply the Dijkstra s algorithm on the path-graphs. In addition, to minimize the usage of the wavelength converters, we set the weight of link between different path-graph layers (i.e. the cost of using a wavelength converter) to be X H so that the cost of using one wavelength converter is larger than using any wavelength-continuous path. According to the way we set the weight of each link, to satisfy a new request, algorithm WAPG first tries to find a wavelength-continuous lightpath (i.e. using the same wavelength on all links along the path) using the lowest indexed wavelength (similar to the FirstFit algorithm) and no wavelength conversion. If it fails, WAPG then tries to find a non wavelength continuous path using minimum number of wavelength converters. The request will be blocked if neither wavelength continuous lightpath (using the same wavelength all along path l) nor non wavelength-continuous lightpath (with help of wavelength conversion) can be found. Note that we can effectively apply algorithm WAPG to the case with sparse wavelength conversion or limited-range wavelength conversion as well. In the case with sparse wavelength conversion only certain selected nodes have links between different layers. On the other hand, in the case with limitedrange wavelength conversion, only limited number of links between different layers exist at every node. It is obvious that in the case without wavelength conversion, no links exist between different layers, in which our algorithm works exactly as the FirstFit algorithm. We compare our proposed algorithm with FirstFit and RandomFit algorithms. The FirstFit algorithm tries to use the first available wavelength-continuous path. On the other hand, if such a wavelength-continuous path is not found, it then assigns the first available wavelength to the first link of the path, for example λ i. On the next link, only if λ i is not available, the first available wavelength, for example λ j (i j) is chosen and a wavelength converter is employed to convert wavelength λ i to λ j, this process continues until a wavelength has been assigned to all the links along the path. Similarly, RandomFit algorithm randomly allocates wavelengths to satisfy the new connection request. B. Numerical Results In this section, we conduct simulations to compare the performance of the proposed algorithm with FirstFit and RandomFit in the USAnet topology with 46 nodes and 76 links. We assume that the traffic is uniformly distributed to all node pairs according to a Poisson process and the holding time is Exponentially distributed. We also assume that every link has one bi-directional fiber, each fiber has 20 bands and each band has 4 wavelengths. Thus the total number of wavelengths on a link is set to X =80. Due to the dynamic nature of the traffic (i.e. connections are established and released dynamically), it does not make sense to compare different algorithms in terms of port count reduction or to assess the benefits of wavelength conversion in reducing the port count. Instead, we will use blocking probability and the maximum number of used wavelength converters at any given time as the performance metrics. Our experiments indicate that it is unnecessary to equip every node with maximum number of full wavelength converters. For the USAnet network with an average load of for each node pair and maximum number of ports (i.e. β =1), our study shows that our algorithm and FirstFit achieve the same blocking performance when the number of converters per link (i.e. d) is bigger than 10 as in the case with maximum number of full wavelength converters (i.e. d =80wavelength converters per link). Figure 10 shows the maximum number of used wavelength converters and Figure 11 shows the blocking probabilities of the heuristics when we set the ratio of bands that can be demultiplexed to wavelengths using BTW ports to be β =0.75. Note that such a limited number of BTW ports may also cause request blocked if wavebanding is not considered properly. We can see that RandomFit is not suited for networks with MG-OXCs as it assigns wavelength randomly and consumes a large number of wavelength converters as shown in Figure 10, which results in inefficient usage of the limited number of ports in MG-OXCs and high blocking probability. More specifically, the inefficient usage of the limited ports comes from two aspects. One is that the random wavelength assignment does not take waveband grouping into

11 11 consideration. The other is that wavelength conversion can only happen at the WXC layer, which means the fiber carrying the wavelength(s) has to be demultiplexed into bands, and then into wavelengths, thus consuming resources (e.g. ports and multiplexers/demultiplexers) in the MG-OXCs, and resulting in poor blocking performance. Fig. 10. Max number of used wavelength converters number of ports and the number of wavelength converters are limited. V. CONCLUSION Waveband Switching (WBS) in conjunction with Three- Layer or Single-Layer multi-granular optical cross-connect (or MG-OXC) architectures has been proposed to reduce the increasing costs and complexity in optical networks. In this paper we have conducted a quantitative performance comparison of the two architectures with dynamic traffic. For the on-line case with dynamic incremental traffic, we have proposed online Integer Linear Programming (On-ILP) models as well as a heuristic called Maximum Overlap Ratio (MOR), which is shown to be better than On-ILP and other heuristic algorithms. We have also shown that using Three-Layer MG-OXCs is better than using Single-Layer MG-OXCs in that the former results in a lower request blocking probability given the same number of ports and traffic load. In the case with fully dynamic traffic, we have developed an efficient heuristic algorithm which uses wavelength converters and group wavelengths to band efficiently, thus achieving good blocking performance in waveband switching networks with wavelength conversion. Through extensive simulations and analysis, we have shown that our algorithm is significantly better in terms of minimizing the number of used wavelength converters and outperforms RandomFit and FirstFit in terms of blocking probability. Our study has also shown the trade-offs between wavelength-hop usage and OXC sizes as well as the performance effect of the percentage of bands (β) that can be demultiplexed into wavelengths in WBS networks. REFERENCES Fig. 11. Blocking performance However, FirstFit is very likely to assign wavelengths to lightpaths sequentially, which helps in wavebanding and thus reducing the number of used ports and blocking probability, but it does not minimize the number of wavelength converters in case they are needed. In fact, FirstFit still consumes a significant number of wavelength converters as shown in Figure 10, which in turn consumes ports and hurts its blocking performance. Since the algorithm tries to use a minimal number of wavelength converters while assigning wavelength sequentially, it performs better than FirstFit and much better than RandomFit, and is especially useful when both the [1] M. Lee, J. Yu, Y. Kim, C. Kang, and J. Park, Design of hierarchical crossconnect WDM networks employing a two-stage multiplexing scheme of waveband and wavelength, IEEE J. on Selected Areas in Communications, vol. 20, no. 1, pp , Jan [2] L. Noirie, M. Vigoureux, and E. Dotaro, Impact of intermediate grouping on the dimensioning of multi-granularity optical networks, in Proceedings of OFC, 2001, p. TuG3. [3] X. Cao, Y. Xiong, V. Anand, and C. Qiao, Wavelength band switching in multi-granular all-optical networks, in SPIE s Proc. vol. 4874, OptiComm 02, Boston Massachusetts, 2002, pp [4] K. Harada, K. Shimizu, T. Kudou, and T. Ozeki, Hierarchical optical path cross-connect systems for large scale WDM networks, in Proceedings of OFC, 1999, p. WM55. [5] O. Gerstel, R. Ramaswami, and W. Wang, Making use of a two stage multiplexing scheme in a WDM network, in Proceedings of OFC, 2000, p. ThD1. [6] E. Ciaramella, Introducing wavelength granularity to reduce the complexity of optical cross connects, IEEE Photonics Technology Letters, vol. 12, no. 6, pp , Jun [7] R. Lingampalli and P. Vengalam, Effect of wavelength and waveband grooming on all-optical networks with single layer photonic switching, in Proceedings of OFC, 2002, p. ThP4. [8] S. Varma and J. Jue, Protection in multi-granular waveband networks, in Proceedings of IEEE GLOBECOM 04, 2004, pp [9] L. Noirie, F. Dorgeuille, and A. Bisson, gbit/s DWDM metropolitan network demonstration with 10 waveband- ADMs and 155 km teralight metro fiber, in Proceedings of OFC, 2002, p. ThH4. [10] R. Izmailov, S. Ganguly, Y. Suemura, I. Nishioka, Y. Maeno, and S. Araki, Waveband routing in optical networks, in Proceedings of IEEE ICC 02, 2002.

12 12 [11] L. Chen, P. Saengudomlert, and E. Modiano, Optimal waveband switching in wdm networks, in Proceedings of IEEE ICC 04, 2004, pp [12] M. Li, W. Yao, and B. Ramamurthy, Same-destination-intermediate grouping vs. end-to-end grouping for waveband switching in wdm mesh networks, in Proceedings of IEEE ICC 05, 2005, pp [13] R. Izmailov, S. Ganguly, V. Kleptsyn, and A. Varsou, Non-uniform waveband hierarchy in hybrid optical networks, in Proceedings of IEEE INFOCOM 03, vol. II, 2003, pp [14] X. Cao, V. Anand, and C. Qiao, Waveband switching in optical networks, IEEE Communications Magazine, vol. 41, no. 4, pp , Apr [15] S. Yao and B. Mukherjee, Design of hybrid waveband-switched networks with oeo traffic grooming, in Proceedings of OFC, 2003, p. WH3. [16] K. Zhu, H. Zang, and B. Mukherjee, A comprehensive study on nextgeneration optical grooming switches, IEEE J. on Selected Areas in Communications, vol. 21, no. 7, pp , Sep [17] P. Bullock, C. Ward, and Q. Wang, Optimizing wavelength grouping granularity for optical Add-Drop network architectures, in Proceedings of OFC, 2003, p. WH2. [18] G. Huiban, S. Perennes, and M. Syska, Traffic grooming in WDM networks with multi-layer switches, in Proceedings of IEEE ICC 02, [19] X. Cao, V. Anand, Y. Xiong, and C. Qiao, A study of waveband switching with multi-layer multi-granular optical cross-connects, IEEE J. on Selected Areas in Communications, vol. 21, no. 7, pp , Sep [20] R. Douville et al., Extensions to generalized MPLS in support of waveband switching, draft-douville-ccamp-gmpls-wavebandextensions-03.txt, Feb [21] E. Dotaro et al., Optical multi-granularity architectural framework, draft-dotaro-ipo-multi-granularity-02.txt, Aug [22] Y. Xin, G. N. Rouskas, and H. G. Perros, On the physical and logical topology design of large-scale optical networks, IEEE/OSA J. of Lightwave Technology, vol. 21, no. 4, pp , Apr [23] H. Zang, J. P. Jue, and B. Mukherjee, Review of routing and wavelength assignment approaches for wavelength-routed optical WDM networks, Optical Networks Magazine, vol. 1, no. 1, pp , [24] X. Yang and B. Ramamurthy, Dynamic Routing in Translucent WDM Optical Networks: The Intradomain Case, IEEE/OSA J. of Lightwave Technology, vol. 23, no. 3, pp , Mar [25] S. Subramaniam, M. Azizoglu, and A. K. Somani, All-optical networks with sparse wavelength conversion, IEEE/ACM Tran. on Networking, vol. 4, pp , Aug [26] M. Kovacevic and A. Acampora, Benefits of wavelength translation in all-optical clear-channel networks, IEEE J. on Selected Areas in Communications, vol. 14, no. 5, pp , Jun [27] B. Ramamurthy and B. Mukherjee, Wavelength conversion in WDM networking, IEEE J. on Selected Areas in Communications, vol. 16, no. 7, pp , Sep [28] E. Karasan and E. Ayanoglu, Effects of wavelength routing and selection algorithms on wavelength conversion gain in WDM optical networks, IEEE/ACM Tran. on Networking, vol. 6, no. 2, pp , Apr [29] X. Cao, V. Anand, and C. Qiao, Multi-layer versus single-layer optical cross-connect architectures for waveband switching, in Proceedings of IEEE INFOCOM 04, vol. III, 2004, pp Vishal Anand (S 99-M 03) received his M.S. and B.E. degrees in Computer Science & Engineering from the University at Buffalo (SUNY), Buffalo, NY and the University of Madras, Madras, India, in 1999 and Subsequently he received his Ph.D. from Buffalo in September 2003, with a concentration in optical telecommunication networks. He is the winner of the North America OPNET Scholarship from OPNET Tech. Inc. and the Outstanding Ph.D. Student award at Buffalo. He received the SUNY Chancellors Promising Inventor Award for his research work in the area of optical networking. He has worked as a research scientist at Bell Labs, Lucent technologies and Telcordia Technologies (ex-bellcore), where he investigated issues relating to traffic routing and survivability in optical networks. His research interests relate to broadband communication networks (IP, ATM, SONET, WDM), optical burst switching and include network planning, modeling, internetworking, QoS and network protocols. He is also interested in Wireless Ad hoc and Sensor networks and issues relating to network security. He has more than 30 peer-reviewed papers in leading technical journals and conference proceedings. Presently he is an Assistant professor at SUNY College at Brockport, Brockport, NY. Chunming Qiao (S 89-M 92) is a Professor at University at Buffalo (SUNY). He directs the Lab for Advanced Network Design, Evaluation and Research (LANDER). He pioneered the research on optical burst switching (OBS) and integrating cellular and ad hoc relaying technologies. His research has been supported by a number of NSF grants including two ITR awards, and by seven other major networking R&D organizations. Dr. Qiao has published more than seventy papers in leading technical journals and 120 conference proceedings, authored eight book chapters, and given several keynote speeches, tutorials and invited talks. He is an editor of several journals and magazines (including ToN), and a guest editor for several others including JSAC and MONET. In addition, he has chaired or co-chaired several international conferences and workshops. networks. Xiaojun Cao (S 01-M 04) received his B.S. degree from Tsinghua University, Beijing, China, in 1996, M.S. degree from Chinese Academy of Sciences in 1999, and Ph.D. degree from The State University of New York at Buffalo in He is currently an assistant professor with the Department of Networking, Security, and Systems Administration at Rochester Institute of Technology. Dr. Cao received the NSF CAREER award in His research interests include modeling, analysis, and protocols/algorithms design for optical and wireless