THE DEMAND for more bandwidth is steadily increasing

Transcription

1 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 8, OCTOBER Routing Framework for All-Optical DWDM Metro and Long-Haul Transport Networks With Sparse Wavelength Conversion Capabilities Ala I. Al-Fuqaha, Member, IEEE, Ghulam M. Chaudhry, Senior Member, IEEE, Mohsen Guizani, Senior Member, IEEE, and Miguel A. Labrador, Senior Member, IEEE Abstract In this paper, we propose a novel routing framework for all-optical dense wavelength-division-multiplexing transport networks with sparse wavelength conversion capabilities. The routing framework includes an integer linear programming formulation to handle the static lightpath establishment problem and a novel open shortest path first protocol extension that advertises the availability of wavelength usage and wavelength conversion resources. Our routing framework addresses the limitations of the extensions presented in the literature because it also includes: 1) an efficient flooding protocol that is suitable for the dynamic nature of these networks and 2) an efficient route and wavelength computation engine that minimizes connection costs without hindering the blocking probability. Index Terms Fuzzy, integer linear programming (ILP), link-state, link-state advertisements (LSAs), open shortest path first (OSPF), routing and wavelength assignment (RWA), sparse wavelength conversion, update policies. I. INTRODUCTION THE DEMAND for more bandwidth is steadily increasing despite the hard times facing telecommunication equipment manufacturers and carriers. This demand motivated the industrial and research communities alike to believe that removing the electronic components from optical transport networks and using the dense wavelength-division-multiplexing (DWDM) technology are key to upgrade the capacity of these networks. This led these communities to embark on the task of developing high capacity, modular, scalable, and flexible all-optical DWDM transport networks with rich monitoring and management capabilities. However, the realization of these networks requires the introduction of many new protocols and mechanisms to control and manage their resources. Many new protocols and mechanisms have been introduced toward this end. However, these protocols were introduced at a very fast pace and they still need to be evaluated, refined and Manuscript received September 1, 2003; revised December 1, A. I. Al-Fuqaha is with LAMBDA Optical Systems, Inc., Reston, VA USA ( gmpls@yahoo.com). G. M. Chaudhry is with the School of Computing and Engineering, University of Missouri, Kansas City, MO USA ( chaudhryg@umkc.edu). M. Guizani is with the Computer Science Department, Western Michigan University, Kalamazoo, MI USA ( mguizani@cs.wmich.edu). M. A. Labrador is with the Computer Science and Engineering Department, University of South Florida, Tampa, FL USA ( labrador@ csee.usf.edu). Digital Object Identifier /JSAC sometimes reinvented. The premise of generalized multiprotocol label switching (GMPLS) is to provide a common control plane (signaling and routing) for networks comprised of devices that switch in different domains: packet, time, wavelength, and fiber. At first glance, this approach might seem to reduce much of the network complexities by eliminating many of the control plane protocols that are currently in use and replacing them with a common control plane (i.e., GMPLS). On the contrary, we believe that this approach will result in having a single but very complex control plane (i.e., based on GMPLS) that tries to be generic enough to deal with the different switching technologies but fails to deal with some of the issues that are particularly important to some of these technologies. While all-optical DWDM transport networks offer new capabilities, several challenges are introduced beyond those known in traditional electro-optical networks. In this paper, we introduce a framework to handle connection routing and wavelength assignment operations in such networks. In the next section, we provide the motivation for a new transport network architecture. Section III provides an introduction to the routing and wavelength assignment (RWA) problem in all-optical DWDM networks with and without the lambda continuity constraint. Section IV provides an integer linear programming (ILP) formulation for the RWA problem in DWDM networks with sparse wavelength conversion capabilities. Section V presents an extension to the open shortest path first (OSPF) routing protocol that enables it to handle routing in all-optical DWDM networks regardless of their wavelength conversion capabilities. Section VI presents two new link-state origination policies to advertise the availability of the wavelength and conversion resources throughout the optical network domain. Section VII introduces a route computation engine based on fuzzy logic that efficiently selects paths for connection requests in optical networks given the wavelength conversion constraints that might exist at each node. Simulation results are presented in this section to compare our proposed fuzzy heuristic with other approaches used in the literature. Finally, Section VIII discusses our findings and future extensions of this work. II. NEED FOR A NEW TRANSPORT NETWORK ARCHITECTURE A quick review of the overlay-based architecture of today s transport networks reveals the inefficiencies associated with them. Fig. 1(a) depicts this architecture, its overlays, and the data rates used at the interfaces between these overlays. Each /04$ IEEE

2 1444 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 8, OCTOBER 2004 Fig. 1. (a) Conventional transport networks: Each signal is terminated at each B-DCS and ADM node. (b) Future transport systems with all-optical OTS systems, the released capacity on the B-DCS and ADM nodes can be used to extend the life of the network. overlay is faster than the overlay on top. This architecture has many disadvantages. A lot of inefficiencies are introduced by the multiple overlays, for example the protocol overhead to carry Internet protocol (IP) traffic over asynchronous transfer mode (ATM) over synchronous optical network (SONET) over DWDM totals to 22% of the bandwidth. Overlays do not often work in concert; for example, every overlay runs at its own speed resulting in low-speed devices not being able to fill up the wavelength bandwidth. Also, when detecting failures, overlays compete to perform protection. The optical transport systems (OTSs) in the optical overlay do not provide automated provisioning resulting in an architecture that does not scale well with increasing demand. SONET ADMs are advantageous for constant bit rate traffic but introduce undesirable latencies for bursty traffic. Another drawback of the traditional overlay-based architecture is that SONET ADMS are inflexible and costly since handling a higher level optical signal (e.g., OC-768) means replacing the current SONET-based equipment with a new one. Also, in this architecture, each switch that demultiplexes the signals will need an electrical network element for each channel even if the traffic on that channel is not dropping at the site. For all these reasons, another architecture that minimizes the number of overlays in the transport network is needed. In order to overcome these inefficiencies, the industrial and research communities believe that the introduction of an overlay of OTSs that support automated circuit provisioning and is transparent to transmission signal formats and data rates is key in order to upgrade the capacity of today s transport networks. Fig. 1(b) illustrates this new architecture. This new architecture allows IP-based services to be carried directly over this new overlay reducing all the inefficiencies associated with transporting these services over frame-relay over ATM over SONET over DWDM. Since Internet traffic projections expect

3 AL-FUQAHA et al.: ROUTING FRAMEWORK FOR ALL-OPTICAL DWDM METRO AND LONG-HAUL TRANSPORT NETWORKS 1445 that IP-based services will constitute a large portion of future Internet traffic, this architecture offers huge advantages and several cost-effective service offerings and features such as automated circuit provisioning, transparency to signal formats and data-rates, unprecedented capacity offerings, protection, and efficient transport of IP-based services. III. RWA IN NETWORKS WITH SPARSE WAVELENGTH CONVERSION CAPABILITIES One of the most important goals in this new transport network architecture is the support of real-time provisioning. Currently, service providers have to go through a lengthy and tedious manual process in order to satisfy a client s request to establish a lightpath. The realization of this vision, however, depends on many factors. Given a request for an optical channel, the RWA problem must be solved so that a route and a wavelength or lambda is assigned to each request from source to destination. The RWA is an NP-complete problem that is usually divided to the more manageable routing and wavelength assignment subproblems. For the routing subproblem, three main approaches are known: fixed routing, adaptive routing, and semi-adaptive routing [3], [5], [10], [11], [14]. For the wavelength assignment subproblem, several heuristics have been proposed such as first-fit, MAX-SUM, least loaded, most used, relative capacity loss, distributed relative capacity loss, and many others [14]. In addition, total cost-based selection, balanced cost-based selection, and future cost-based selection heuristics that solve both problems in an integrated manner have been recently introduced in [12]. In the absence of wavelength converters, a lightpath must be established from a source to a destination using the same wavelength (lambda); this is a wavelength constraint network. On the other hand, with wavelength converters, lightpaths can be converted to different wavelengths, leading to an expected lower call blocking probability. Lightpath requests are commonly classified as static or dynamic. With static requests, users demands are known in advance and the provisioning problem is to set up lightpaths to satisfy all the requests, while minimizing the amount of network resources, such as the number of wavelengths and wavelength converters used by the lightpath. In this case, the RWA problem is known as the static lightpath establishment (SLE) problem. With dynamic requests, connections are requested in a dynamic fashion and the RWA problem will try to establish lightpaths that minimize the blocking probability. This is known as the dynamic lightpath establishment (DLE) problem. Optical switches can be of two different types according to their conversion capabilities. Full wavelength conversion switches are those that can convert an incoming wavelength to any outgoing wavelength and in addition, the number of converters is equal to the total number of outgoing wavelengths. On the other hand, partial wavelength conversion switches are equipped with an optimal number of wavelength converters to minimize the high cost of these devices. Several studies have already considered the case of networks with wavelength converters. For example, [8], [12], and [13], consider the case of optimal converter placement, where the switches have full or limited conversion capabilities. Lately, it has been shown that the RWA and switch placement problems, which are usually solved separately, should be considered in an integrated manner [9]. In this paper, we assume sparse and limited wavelength conversion resources. By sparse converter resources, we mean that the optical network might not have enough wavelength resources to satisfy every request requiring wavelength conversions. By limited wavelength converter resources, we mean that the wavelength converter resources installed in the optical network might not have the capability to convert any input wavelength to any output wavelength. It has been shown that networks with sparse and limited conversion capabilities can achieve similar blocking probability in a more cost-effective manner [13] than networks with full wavelength conversion. IV. ILP FORMULATION FOR THE RWA PROBLEM IN NETWORKS WITH SPARSE WAVELENGTH CONVERSION CAPABILITIES (RWA-SWC) In this section, we present an ILP formulation to solve the RWA problem in networks with sparse wavelength conversion capabilities. In this formulation, the goal of the objective function is to minimize the total cost of all lightpaths that need to be established in the optical network. As stated before, this is an NP-complete problem. In order to formulate the problem, let us define the following. : Number of switches. : Number of links. : Number of wavelengths per link. : Total number of lightpaths that need to be established. : Number of source-destination pairs. : Vector of size, where element represents the number of requested lightpaths between the th source-destination. : Vector of size, where element represents the number of all possible paths between the th source-destination pair. : Vector of size, where element represents the number of wavelength converters installed on the th node. : A list of vectors that represent the paths on which each of the source-destination pairs can be routed, is the th vector of the list. Element represents the th path on which the th source-destination pair can be routed. These paths can be enumerated using the -shortest paths algorithm. Notice that two paths are considered to be distinct if they go through different fibers or different wavelengths in their route from source to destination. : A list of matrices that represent the usage of the link resources by the different paths, matrix is the th matrix of the list. Element if the th path between the th source-destination pair uses link, otherwise,. : A list of matrices that represent the usage of the wavelength conversion resources by the different paths, matrix is

4 1446 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 8, OCTOBER 2004 the th matrix of the list. Element if the th path between the th source-destination pair uses a wavelength converter that is installed on node, otherwise,. : A list of vectors that represent the cost of the different paths, vector is the th vector of the list. Element is the cost of the th path between the th source-destination pair. : A list of matrices that represent the usage of the wavelength resources (lambdas) by the different paths, matrix is the th matrix of the list. Element if the th path between the th source-destination pair uses wavelength on link, otherwise,. : A list of vectors that represent the cost of the different paths, vector is the th vector of the list. Element if the th path between the th source-destination pair is selected, otherwise,. The objective function of the RWA problem in networks with sparse wavelength conversion capabilities can be defined to minimize the total cost of establishing all requested lightpaths. The RWA-SWC problem is then formulated as follows: Minimize Subject to the following constraints: (1) (2) (3) (4) (5) In this formulation, the symbol indicates the transpose operation. Equation (1) indicates that a path can be selected or not selected (binary variable). Equation (2) indicates that all the requested lightpaths need to be established for a solution to be feasible. Equation (3) verifies that no more than wavelengths are used on a single link. Equation (4) verifies that the wavelength conversion capability constraints are respected. Finally, (5) guarantees that no more than one connection is carried on any given wavelength of all links in the network. In order to use the formulation presented above on reasonable size networks, we propose a pruning strategy that aims at reducing the search space of possible routes and wavelengths under the scenario of static requests or static lightpath establishment (SLE). This pruning strategy aims at: 1) limiting the possible routes between a given source-destination pair; 2) limiting the possible wavelengths to only those that can be generated by the tunable lasers and wavelength converters technologies installed in the network; 3) limiting the possible wavelengths to be the same before and after nodes that do not support wavelength conversion; 4) limiting the possible wavelengths to a subset of the wavelengths that the DWDM links can support. The reduced search space is then presented to the ILP formulation, which selects the routes and wavelengths that need to be assigned to the given connections. The result is that these connections are routed throughout the optical network with the least possible cost obeying the wavelength conversion restrictions present in the network domain. The difficulty of any ILP problem depends on the number of variables and constraints in that problem. The factors that determine the number of variables and constraints used in the above formulation are the number of connections that need to be established, the number of nodes in the network, the number of links in the network, the number of nodes that possess wavelength conversion resources, the type of wavelength conversion used within the network, the number of paths that need to be considered between a given source-destination pair, and the number of wavelengths that can be carried on a single link. The following two equations provide a simple estimate of the number of variables and constraints involved in the ILP problem: Number of Variables Number of Constraints where number of lightpath requests that need to be established on the network; average number of possible routes that can be considered between a given source-destination pair; number of wavelengths that can be carried over the links of the network; average number of hops used to route a given connection; percentage of wavelength options that can be preeliminated due to technology constrains or due to the lack of wavelength conversion resources; percentage of wavelength options that can be preeliminated due to the user s educated decision that a subset of the supported wavelengths can be used to route all the connections in hand. The above equations also hold for networks with the wavelength continuity constraint by substituting. The process of estimating can sometimes be complicated, in such cases upper and lower bounds can be utilized to get a bounded estimate of the number of variables and constraints involved in the ILP problem. Typical ILP problems found in real-life situations have 2000 variables and 4000 constrains. As a quick rule of thumb, the above formulation will be helpful as long as the number of variables and the number of constraints are around these typical numbers. Figs. 2 and 3 illustrate the relationship between the number of variables and constraints involved in the ILP problem and and, respectively. Notice the effect that and have on reducing the number of variables and constraints involved in the ILP problem. Table I illustrates two simple scenarios to which we applied the ILP formulation presented above. Table I also indicates the optimal resources that need to be allocated to each lightpath.

5 AL-FUQAHA et al.: ROUTING FRAMEWORK FOR ALL-OPTICAL DWDM METRO AND LONG-HAUL TRANSPORT NETWORKS 1447 Fig. 2. Number of variables and constraints versus P1. For lightpaths =20, K =2, W =16, H =1:6, and P2=0:2. Fig. 4. Sample network with SWC capabilities. presented in this section becomes of limited use. In Section VII, we present a RWA strategy to handle such networks. V. OSPF ROUTING EXTENSION IN SUPPORT OF ALL-OPTICAL DWDM NETWORKS WITH SPARSE WAVELENGTH CONVERSION CAPABILITIES Fig. 3. Number of variables and constraints versus P2. For lightpaths =20, K =2, W =16, H =1:6, and P1=0:3. TABLE I EXAMPLES OF ROUTE, WAVELENGTH, AND CONVERTER ASSIGNMENT IN SWC NETWORKS Fig. 4 shows the topology of the network on which the lightpaths indicated in Table I need to be established. In this example, each wavelength converter is assumed to have a cost of 100. We implemented the -shortest paths algorithm to enumerate the different paths for the ILP formulation and then we used CPLEX to solve the formulation. It should be noted here that as the number of variables and constraints involved in the ILP problem grows, the formulation As explained previously, a requested optical lightpath needs to be assigned a route and a single or a set of wavelengths throughout the optical network domain from source to destination. A routing protocol is required to disseminate wavelengths and converters availability within the optical network domain. Kompella and Rekhter presented an Internet draft in which they discussed the information that needs to be flooded by any routing protocol in support of GMPLS [7]. In that draft, a generic approach to handle networks comprised of packet switch capable (PSC), time-division-multiplexing capable (TDMC), lambda switch capable (LSC), and fiber switch capable (FSC) equipment was presented, but the draft did not address routing in networks comprised of LSC switches with any kind of wavelength conversion capabilities. We think that the approach presented in [7] complicates the routing protocol and makes it inefficient to handle LSC switches since it must handle the advertisements of equipment employing all previously mentioned switching technologies even though such equipment might belong to different overlays as explained in Section II. We believe that telecom networks employ an overlay architecture and it is more efficient and feasible to design a routing protocol that is specific to each of the employed overlays. In this case, despite the fact that each overlay would employ its own routing protocol, it would be able to advertise more information that is specific to its intended overlay, resulting in more efficient routing and better provisioning of network resources. In this section, we present an extension to the OSPF protocol that addresses the routing problem faced by all-optical DWDM networks with sparse wavelength conversion resources. Even though the routing extension presented here is an overlay specific one that pertains to all-optical DWDM transport networks [photonic overlay in Fig. 1(b)] regardless of their wavelength conversion capabilities, a similar approach can be taken to design routing protocols for other overlays.

6 1448 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 8, OCTOBER 2004 Fig. 6. Converter availability opaque LSA. Fig. 5. Wavelength availability opaque LSA. Our proposed routing extension is based on three major changes to the OSPF routing protocol. The first change consists of the introduction of two new link-state advertisements (LSAs) to advertise wavelengths and converters availabilities. The second change involves modifications to the flooding policy of the OSPF protocol to make it more suitable for the dynamic nature of all-optical DWDM networks. The last change involves modifications to the route computation component used in the original OSPF protocol that minimizes the cost of the lightpaths, while not hindering the blocking probability. The following two subsections address the first change, while Sections VI and VII address the changes to the flooding policy and the route computation engine, respectively. A. Description of the New LSAs The purpose of the introduced LSAs is to advertise the availability of each wavelength per fiber and the number of wavelength conversion resources available within a switch. Our extension is generic and it can handle all-optical DWDM networks composed of LSC switches regardless of their wavelength conversion capabilities. The introduced LSAs use the OSPF opaque LSA option as a vehicle to advertising these parameters. The OSPF opaque LSA option defined in [4] provides a generalized mechanism for the OSPF protocol to carry additional information. An opaque LSA consists of a standard LSA header followed by 32-bit aligned application specific information field, which is divided into TLV tuples of type, length, and value. In addition to the fields defined in [6] and [7], the wavelength availability opaque LSA includes the following fields (see Fig. 5). : This is a new sub-tlv that we introduce to the OSPF protocol to represent the link protection type. The value of the link protection type can be from 1 to 6 to indicate the link protection type as explained in [7]. : This is a new sub-tlv that we introduce to the OSPF protocol to represent all the Shared Risk Link Groups (SRLGs) to which the link belongs. : This is a new sub-tlv that we introduce to the OSPF protocol to represent the usage profile of the wavelengths carried on the link described in this LSA. Length of Mask: Number of bits used to represent the bandwidth mask. Wavelength Availability Mask: This field represents the usage profile of all wavelengths on a specific link. This field can be extended or shortened as needed using the length of mask field. If the value of the th bit of this filed is set to 1, then this indicates that the th wavelength of the specified link is used. When the bit value is set to 0 this indicates that the wavelength is free and it can be assigned to an incoming lightpath. Fig. 6 depicts the structure of the converter-availability opaque LSA, where the difference between the total and used wavelength converter fields represents the total number of converters that are not used of a given converter type within the switch. This LSA contains the following fields. : We use the same concept used in defining the wavelength availability opaque LSA. We decided to assign a type value of to the converter availability TLV. Converter Type: Different types of wavelength converters are commonly used in all-optical DWDM networks, so this field is used to specify the type of wavelength conversion resources installed in the network. If multiple wavelength conversion types are installed in the network, multiple fields can be used to specify the type and availability of these different types, as shown in Fig. 6. This field enables the protocol to convey important information about the type of wavelength converters installed on a given switch allowing the route and wavelength assignment engine to distinguish between full-range wavenlength converters, limited-range wavelength-converters, and wavelength shifters. Total: The total number of wavelength converters of the specified type that are installed on the switch. Used: The total number of wavelength converters of the specified type that are currently in use. In the TLVs and sub-tlvs presented above, we selected the range from to to represent the link protection type, Shared Risk Link Group (SRLG), wavelength availability, and converter availability TLVs and sub-tlvs, respectively.

7 AL-FUQAHA et al.: ROUTING FRAMEWORK FOR ALL-OPTICAL DWDM METRO AND LONG-HAUL TRANSPORT NETWORKS 1449 Fig. 7. Snapshot from our all-optical network simulation tool showing a typical 16-node long-haul all-optical DWDM network, where each link carries (8) wavelengths. The wavelength and converter availability opaque LSAs introduced above are generic to handle all-optical DWDM networks with different degrees of wavelength conversion capabilities. These LSAs provide a big advantage over the OSPF extension presented in [7] which is not capable to handle LSC or FSC networks with different degrees of wavelength conversion capabilities. To illustrate the importance of the wavelength and converter availability opaque LSAs that we introduced above, we performed a simulation study on the 16-node long-haul all-optical DWDM network illustrated in Fig. 7. We generated random calls with exponentially distributed call holding and call interarrival times. The source and destination nodes of the generated calls were selected based on a uniform distribution. Also, the route and wavelength assignments for the generated calls were based on the least-cost and most contiguous route and wavelength assignment heuristics. The results of our study are depicted in Figs Fig. 8 compares the degree of blocking perceived when the wavelength and converter availability LSAs are exchanged between the switches with that perceived when no wavelength and converter availability LSAs are exchanged. It is clear from this figure that the exchange of wavelength and converter availability LSAs between the switches in the network can drastically decrease the degree of blocking perceived in the network. Fig. 9 depicts the average number of retries (or call crankbacks) that need to be made before a call can be established when no wavelength or converter availability LSAs are exchanged. Call crankbacks can result in a drastic increase in the network offered load. These crankbacks or retries can Fig. 8. Call blocking probability versus traffic load with and without advertisements assuming 50% of wavelength conversion capability. almost be eliminated when the switches exchange the wavelength and converter availability LSAs. This is possible since the exchanged LSAs inform the switches about the availability of the wavelength and converter resources throughout the network. So, when a node runs the RWA algorithm for a call and informs the signaling mechanism to set up the end-to-end path, the probability that the selected resources will not be in use is very high. Fig. 10 compares the degree of blocking perceived when the wavelength and converter availability LSAs are exchanged between the switches with that perceived when no wavelength and converter availability LSAs are exchanged in networks with different degrees of wavelength conversion resources. We define the degree of wavelength conversion here to be the ratio of the

8 1450 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 8, OCTOBER 2004 Fig. 11. IOA policy. Fig. 9. Average number of call retries versus traffic load with and without advertisements assuming 50% of wavelength conversion capability. Fig. 10. Call blocking probability versus degree of wavelength conversion with and without advertisements. number of wavelength conversion resources installed in the network to the total number of wavelengths in the network. Fig. 10 shows that the wavelength and converter availability LSAs can greatly enhance the blocking performance of the network especially when the degree of wavelength conversion in the network is low. VI. MODIFIED OSPF ORIGINATION POLICY The original OSPF standard defines a mechanism to originate the router and network LSAs. Similarly, a mechanism needs to be defined to originate the wavelengths and converters availability LSAs that we defined in Section V. It should be emphasized here that the OSPF extensions in support of GMPLS presented in [6] do not address this aspect of the protocol. In this section, we propose two different origination policies. The first one is simple and serves as our base policy for comparison. The second policy is very efficient in handling the dynamic nature of all-optical DWDM networks with different degrees of wavelength conversion. Several link-state update policies have been proposed in the literature to minimize the routing protocol overhead needed to exchange the quality-of-service (QoS) parameters needed for QoS routing. A link-state update policy determines when a node should originate link-state update messages and the contents of these updates. Apostolopoulos et al. [1] classified the mechanisms used to trigger the link-state update messages into threshold, class, and timer-based trigger policies. The tradeoffs between these mechanisms lie in the volume of link-state update messages and the accuracy of the state information available to the route computation engine. The exchange of the link-state information at a higher rate results in more accurate state information provided to the route computation engine. This means that the route computation engine will be able to provide lower call blocking probability at the expense of a large volume of link-state traffic. Similarly, exchanging the link-state information at a lower rate provides the route computation engine with less accurate information about the state of the network. This means that the route computation engine will encounter a higher degree of call blocking but at the same time, the network control plane is not overwhelmed with large volume of link-state updates. However, the heuristics presented in [1] have been applied to IP-based networks. In this section, we propose two link-state update mechanisms for dynamic all-optical DWDM networks with different degrees of wavelength conversion. A. Immediate Origination Approach (IOA) Using this link-state update policy, each node should originate wavelength-availability and converter-availability opaque LSAs whenever a new router LSA is originated. Also, each node should originate a wavelength-availability opaque LSA for each of its outgoing links whenever the wavelength availability mask of the link is changed. Moreover, each link should originate a converter availability LSA whenever the usage profile of the wavelength conversion resources installed on the switch are changed. Fig. 11 depicts this simple origination mechanism that advertises the wavelengths and converters availability LSAs as soon as the availability profiles of these resources change. We call this approach the IOA. The strength of this approach stems from its simplicity as it can be used in long-haul DWDM all-optical networks where the lightpath requests are static and do not change frequently. However, in networks with dynamic lightpath requests, this approach can result in a large volume of link-state updates since the availability of the wavelength and conversion resources is constantly changing. Since the lightpath requests presented to access and metro-edge all-optical DWDM networks are usually dynamic, the immediate link-state update approach presented in this section is not efficient in handling such networks. The following

9 AL-FUQAHA et al.: ROUTING FRAMEWORK FOR ALL-OPTICAL DWDM METRO AND LONG-HAUL TRANSPORT NETWORKS 1451 Fig. 12. Fuzzy rule base of proposed link-state origination policy. section presents another link-state update mechanism that utilizes fuzzy logic. This mechanism can efficiently handle networks with dynamic lightpath requests, as it is the case in most access and metro-edge all-optical networks. B. Fuzzy Origination Approach (FOA) Since the availability of wavelength and wavelength-conversion resources of all-optical DWDM networks installed in metro-edge and metro-core environments change frequently, a smart link-state update policy that minimizes the exchange of link-state information, while not hindering the blocking probability is needed. In order to satisfy this requirement, in the following subsections we introduce the fuzzy logic-based origination approach (FOA). 1) Fuzzy Inference System: Our fuzzy-based link-state update policy is based on two simple rules, as shown in Fig. 12. The first rule causes the link-state information to be exchanged less often when the wavelength and wavelength-conversion resources installed in the network are lightly utilized. While the second rule causes the link-state information to be exchanged more often when the wavelength or wavelength-conversion resources installed in the network are highly utilized. The rationale behind these rules is very simple. In the first case, inaccuracies in the state information pertaining to the availability of the wavelength and wavelength-conversion resources in the network do not increase the call blocking probability because old information related to resources not utilized is still valid. In the second case, the network becomes more utilized and, therefore, most of its resources experienced changes since the last update, augmenting the error in the state information. Therefore, in order to avoid increasing the call blocking probability unnecessarily, state information must be advertised more frequently. The net effect of our policy is the reduction of the volume of state information exchanged, while not increasing the call blocking probability. The fuzzy interference system presented in Fig. 12 is based on the linguistic approach that depends on linguistic variables whose values are words or sentences in a natural or artificial language rather than numbers. Our fuzzy link-state update policy utilizes human expert experience to identify when the network resources in terms of wavelengths and converters are considered to be highly utilized or lightly utilized. However, since the imprecision or fuzziness is inherent in human judgments, representing the utilization of the different resources using linguistic variables makes it easier and more flexible for the human operator to specify the level of usage of the different network resources. Then, a fuzzy-inference rule base can be used to aggregate the network resource utilizations in terms of wavelength and converter resources into a single value that specifies the waiting factor to send the next link-state update message. The waiting factor calculated using the fuzzy-inference rule base is an absolute number. This number should be multiplied by the average call interarrival time to calculate the actual waiting time between two consecutive link-state updates. No link-state update needs to be originated when the calculated waiting time between updates expires before having a new link-state update. The fuzzy link-state update policy presented in this section can be easily extended or modified since it is based on common sentences (rules). On the other hand, it is very difficult to represent these rules using mathematical functions. But even if it is feasible to do that, it is definitely not easy to modify them as frequently as required. Even though the link-state update policy presented here applies to all-optical DWDM networks with sparse wavelength conversion capabilities, a similar approach can be used to design a fuzzy-inference rule base that applies to IP-based networks and advertise QoS parameters. 2) Membership Functions: The proposed FOA model utilizes two linguistic variables to represent the availability of the wavelength and converter resources in the optical network and one output linguistic variable to represent the waiting factor between two consecutive link-state updates. The membership functions assigned to these variables are chosen as -functions and -functions for the input linguistic variables and Gaussian functions for the output variable. The and membership functions are chosen for the input variables because they can be configured to provide fast as well as slow transitions from membership to nonmembership and vice versa (e.g., from high-degree of utilization to a low-degree of utilization). The Gaussian membership function is chosen for the output variable because it results in smooth switching as the input linguistic variables change values. It is worth mentioning here that studies conducted on fuzzy systems have shown that the choice of membership functions does not drastically change the behavior of the system. The and membership functions are specified by two parameters and and the Gaussian membership function is specified by two parameters and as follows: Gaussian Fig. 13 shows the general form of the resource availability membership functions. Low membership function provides different degrees of membership that range from full membership when the resource utilization is lower than 10% to nonmembership when the resource utilization exceeds 70%. On the other hand, High membership functions provides different degrees

10 1452 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 8, OCTOBER 2004 Fig. 13. Resource availability input membership functions. Fig. 14. Waiting factor output membership functions. of membership that ranges from nonmembership when the resource utilization is lower than 30% to full membership when the resource utilization exceeds 90%. A human operator can easily tune the parameters involved in these membership functions. Fig. 14 shows the general form of the waiting factor output membership functions. Low waiting factor means that the waiting time between two consecutive link-state originations is low resulting in more frequent updates. While high waiting factor means that the waiting time between the link-state updates is longer resulting in less frequent exchange of link-state update messages on the network control plane. The output membership functions have the following parameters: Low Waiting Factor Gaussian High Waiting Factor Gaussian Fig. 15 provides an example that illustrates the computation of the update interval based on the fuzzy rule base. The recommended rules try to maximize the waiting factor as long as the wavelength and converter resources in the network are not highly utilized. Higher resource utilizations result in lower waiting factors. The waiting factor increases gradually as the resource utilizations increase. This helps in minimizing the linkstate update messages exchange on the optical network control plane whenever that is possible. It is worth noticing that our policy utilizes the min, max, min, max, and centroid methods Fig. 15. Example of applying our fuzzy inference system rules to calculate the interval between link-state updates. for the fuzzy and, or, implication, aggregation, and defuzzification operators, respectively. C. Performance Results We carried out a performance study of the IOA and FOA approaches. Fig. 16 plots the average number of update messages exchanged under different traffic loads. The figure shows that the average number of messages exchanged using our FOA strategy is considerably smaller than the one needed by the IOA LSA origination policy. In addition, in Fig. 17, we compare the blocking probability of both schemes under different traffic loads and show that the FOA strategy does not increase the blocking probability. Based on these findings, we conclude that

11 AL-FUQAHA et al.: ROUTING FRAMEWORK FOR ALL-OPTICAL DWDM METRO AND LONG-HAUL TRANSPORT NETWORKS 1453 Fig. 16. Number of link-state updates exchanged versus traffic load for the immediate and fuzzy-based link-state origination policies. Fig. 17. Call blocking probability versus traffic load for the immediate and fuzzy-based link-state origination policies. our fuzzy-based LSA origination policy enables the routing protocol to exchange less update messages without degrading the blocking performance. VII. PROPOSED RWA STRATEGIES Now, we apply fuzzy logic to routing in all-optical DWDM networks with sparse wavelength conversion capabilities. In the proposed approach, a fuzzy-inference rule base is used to assign a fuzzy cost to each path based on crisp metrics that reflect the availability of resources within the network. The fuzzifier module takes these crisp inputs and generates fuzzy values that can be used by the fuzzy inference system. Next, the fuzzy inference system applies the rules available in its rule base to generate a fuzzy cost for each of the possible paths. Then, the defuzzifier module takes the fuzzy cost and generates a crisp cost for each of the possible paths. Finally, the path selection module compares the costs of all the possible paths and selects the best possible route. The proposed fuzzy-based model is shown in Fig. 18. In this model, the -shortest paths module is used to find the best possible routes for the requested lightpath. At the same time, the model monitors the availability of the wavelength and converter resources within the network domain by accessing the routing protocol link-state database (LSDB) that records the wavelength and converter LSAs described in Section V. The wavelength assignment module employs a simple heuristic that we developed called the most contiguous wavelength assignment heuristic. In this heuristic, a set of wavelengths is assigned to the route in order to minimize the use of the wavelength converters. Our most contiguous wavelength assignment heuristic works by choosing the wavelength that is most contiguous (avoiding wavelength conversion) and uses wavelength conversion when the rest of the path cannot continue on the same wavelength (wavelength is used). The rest of this section is organized as follows. Section VII-A introduces our fuzzy-based route selection module, the designed rule base, and the membership functions. Section VII-B, introduces our most contiguous wavelength assignment heuristic. Finally, Section VII-C provides our simulation results comparing the performance of our proposed fuzzy-based route selection and most contiguous wavelength assignment heuristic with some of the well known and used approaches presented in the literature. A. Fuzzy Route Selection Module In all-optical DWDM networks, it may be too simplistic to employ conventional routing strategies that are based on the evaluation of a single routing metric. In such networks, it is crucial to employ routing algorithms that provide optical paths with the lowest possible cost, while maintaining a low lightpath blocking probability at the same time. Additionally, these routing algorithms should avoid the usage of the wavelength conversion resources installed within the optical network. Given these requirements and the complex tradeoffs between them, it is difficult to define a single routing metric for routing algorithms in such networks. Therefore, a new routing paradigm that searches for acceptable routes for intended lightpaths, while satisfying their QoS requirements is required for all-optical DWDM networks. Such a paradigm will affect not only the QoS offered to the optical lightpaths but also the utilization of the network resources and the blocking probability encountered in the optical network. In this section, we propose a fuzzy-based heuristic for the routing problem in DWDM networks with sparse wavelength conversion capabilities. The proposed approach employs a fuzzy-inference rule base to assign a fuzzy cost to each path based on the crisp metrics of the path, network links, and network resource utilization (wavelength and converter resource utilizations). 1) Fuzzy Inference System: Routing algorithms are needed to select a set of links that need to be used to route the lightpath through the optical network from its source to its destination. The problem of assigning a route to an optical lightpath based on two or more QoS parameters is an NP-complete problem. In this section, we propose using fuzzy logic techniques to solve this problem. The challenge of this work is to route lightpaths through the optical network over paths that satisfy the QoS requirements of the routed lightpaths, without hindering the blocking probability encountered throughout the optical network. Our fuzzy-based route selection module is based on 12 simple rules, as shown in Fig. 19. The recommended rules try to assign a fuzzy cost for each path based on the QoS parameters of that path as well as on global network state information. When network bandwidth congestion level is high and network converter congestion level is low, shorter paths are more preferred even if they use more wavelength converter resources since these resources are not heavily utilized. In a similar manner, when network bandwidth congestion level is low and network converter

12 1454 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 8, OCTOBER 2004 Fig. 18. Fuzzy-based routing model for all-optical DWDM networks with limited wavelength conversion capabilities. congestion level is high, paths that use less wavelength converter resources are preferred even if they are longer in terms of their number of hops since the network links are not highly utilized. This helps in assigning fuzzy costs to paths based on the network and path state information in such a way that conserves the network resources when these resources are highly utilized and at the same time provides lower cost paths when the network resources are not highly utilized. The fuzzy route selection module presented in this section can be easily extended or modified since it is based on common sentences (rules). A similar approach can be used to design a fuzzy-inference rule base that applies to IP-based networks to select paths based on multiple QoS parameters. 2) Membership Functions: In our fuzzy route selection module, membership functions are used in the antecedents and consequents of rules. The proposed model utilizes eight linguistic variables to represent the route QoS metrics and the network state information. The model also has one output linguistic variable that represents the overall fuzzy cost assigned to the path. The membership functions assigned to these variables are chosen as -functions and -functions for the input linguistic variables and Gaussian functions for the output variable. In the following, we provide the general form and parameters of the eight input membership functions used in our fuzzy route selection module. Hops-high: This membership function is -shaped (see Fig. 20), the values of its and parameters depend on the spacing between the dispersion compensation modules (DCMs), amplifiers, switching systems noise figures, and fiber lengths in the network. We suggest the following strategy to assign values for the and parameters. When the optical signals exchanged between the different source-destination pairs in the network have lowpower levels or optical signal to noise ratios (OSNRs), the values assigned to the and parameters should be low (e.g., and ). When the optical signals exchanged between the different source-destination pairs in the network have high-power levels and OSNRs, the values assigned to the and parameters can be high (e.g., and ). Cost-high: This membership function is -shaped, the values of its and parameters depend on the range of values used for the cost metric. Usually, the cost metric is a number between 0 and 255. In this case, we suggest that the values of the and parameters be and. Network bandwidth congestion-high: This membership is -shaped with parameters and. The value of the parameter can be lowered to assign higher overall cost for connections that use more hops when the network bandwidth congestion level is high. This helps penalize connections that use more hops when the network bandwidth conversion level is high. Network converter congestion-high: This membership is -shaped with parameters and. The value of the parameter can be lowered to assign higher overall cost for connections that use more wavelength converters when the network converter congestion level is high. This helps penalize connections that use more wavelength conversion resources when the network converter conversion level is high. Diversity-low: This membership is -shaped (see Fig. 21); the value of the parameter can be increased in networks with redundant links. This helps penalize connections that use unprotected paths, thus encouraging the connection request to be routed over protected paths. Path bandwidth congestion-high: This membership function is similar to the Network bandwidth congestion-high membership function. Choosing the right value for the parameter of this function encourages connection to select paths that are not congested.

13 AL-FUQAHA et al.: ROUTING FRAMEWORK FOR ALL-OPTICAL DWDM METRO AND LONG-HAUL TRANSPORT NETWORKS 1455 Fig. 19. Fuzzy rule base of proposed route selection module. Fig. 20. s-shaped cost-high membership function. Fig. 21. z-shaped diversity-low membership function. Path converter congestion-high: This membership function is similar to the Network converter congestion-high membership function. Choosing the right value for the parameter of this function encourages connection to select paths with less wavelength conversion congestion. Number of converters-high: This membership is -shaped; the value of the parameter can be lowered to assign higher overall cost for connections that use more wavelength converters. This helps penalize connections that use more wavelength conversion resources.

14 1456 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 8, OCTOBER 2004 Fig. 22. Output membership functions used for fuzzy route selection. Fig. 23. Example of applying our fuzzy inference system rules to calculate the overall path cost. Fig. 22 shows the general form of the overall cost output membership functions. The output membership functions have the following parameters: Low Cost Gaussian Low Hops Gaussian Low Converters Gaussian High Diversity Gaussian Low Diversity Gaussian High Converters Gaussian High Hops Gaussian High Cost Gaussian The output membership functions used in our fuzzy route selection model are based on the Gaussian membership function with spacing of 100 units between the functions and. A human operator can adjust these parameters to calculate the overall fuzzy cost of a path in a different way. Fig. 23 provides an example that illustrates the computation of a fuzzy cost for each of the possible routes generated by the -shortest paths module. The recommended rules try to assign fuzzy costs to paths based on the network and path state information in such a way that conserves the network resources when these resources are highly utilized and at the same time provides lower cost paths when the network resources are not highly utilized. It is worth noticing that our policy utilizes the probabilistic or, max, min, max, and centroid methods for the fuzzy and, or, implication, aggregation and defuzzification operators, respectively.

15 AL-FUQAHA et al.: ROUTING FRAMEWORK FOR ALL-OPTICAL DWDM METRO AND LONG-HAUL TRANSPORT NETWORKS 1457 Fig. 24. Proposed most contiguous wavelength assignment heuristic. B. Most Contiguous Wavelength Assignment Heuristic The wavelength assignment problem has been studied extensively [14]. A large number of wavelength assignment schemes have been proposed in the literature, as explained in Section III. However, none of these wavelength assignment schemes account for the scarcity of the wavelength conversion resources available in networks with sparse conversion capabilities. For such networks, we propose a simple wavelength assignment scheme that minimizes the use of wavelength conversion resources as much as possible. The rationale behind this is that the wavelength conversion resources are very scarce in such networks and having a wavelength assignment technique that conserves the usage of these resources is a critical requirement that can drastically conserve the usage of these resources and, thus, enhance the network blocking performance significantly. Fig. 24 provides a high-level description of the proposed algorithm. It should be noticed here that the proposed algorithm conserves wavelength conversion resources as much as possible. However, when a tie occurs between multiple wavelength assignment options, any of the simple pack/spread wavelength assignment schemes presented above can be used to break the tie. We suggest using the first-fit wavelength assignment scheme to break such ties because of the simplicity and good performance

16 1458 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 22, NO. 8, OCTOBER 2004 Fig. 25. Comparison of the average path cost versus degree of wavelength conversion of our fuzzy-routing and most contiguous wavelength assignment heuristic with the least-hops first-fit and least-cost most-used heuristics. of this scheme. Also, notice that the algorithm proposed does not guarantee that it will always find the wavelength assignment with the lowest possible number of wavelength converters. A scheme that will always find the lowest number of wavelength converters can be computationally extensive and the scheme proposed here provides a good balance between simplicity and the efficiency of the found solutions. C. Performance Results The performance of our proposed fuzzy-based routing heuristic has been compared with that of the shortest-path routing and first fit wavelength assignment approach. The lightpath requests presented to the simulated network follow a Poisson arrival process (i.e., the interarrival times are exponentially distributed); the parameter of the arrival process depends on the traffic load presented to the simulated network. The call holding time utilized in this simulation is also exponentially distributed with an average of 15 s. Our simulation tool generated one million lightpath requests to determine the blocking probability of the network. The source and destination of the generated lightpath requests are selected with uniform probability. Figs compare the performance of our proposed fuzzy-routing approach when combined with our proposed most contiguous wavelength assignment heuristic. It is clear that our routing and wavelength assignment approach resulted in better blocking performance and better usage of the wavelength conversion resources, while not compromising the cost (or quality of service) of the selected lightpaths. VIII. CONCLUSION AND FUTURE WORK In this paper, we present a complete framework to handle the static and dynamic lightpath establishment problems in all-optical DWDM network with sparse wavelength conversion capabilities. For the static lightpath establishment, we present an ILP formulation for the RWA problem that applies to networks with different degrees of wavelength conversion capabilities. We also present a pruning strategy that helps to reduce the number of Fig. 26. Comparison of the average number of converters versus degree of wavelength conversion of our fuzzy-routing and most contiguous wavelength assignment heuristic with the least-hops first-fit and least-cost most-used heuristics. Fig. 27. Comparison of the call blocking probability versus degree of wavelength conversion of our fuzzy-routing and most contiguous wavelength assignment heuristic with the least-hops first-fit and least-cost most-used heuristics. variables and constraints of the ILP formulation. We also present an extension to the OSPF protocol in terms of two new opaque LSAs to convey the availability of the wavelength and wavelength-conversion resources within the network. Two new linkstate origination policies are also introduced and their performance is compared. Finally, we present a new wavelength assignment heuristic called the most contiguous wavelength assignment heuristic and a new fuzzy-based route computation engine targeted for all-optical DWDM networks. The performance of our route computation engine is also compared with other heuristics found in the literature and we show that our approach conserves wavelength-conversion resources and finds lower cost paths without hindering the call blocking probability. In the future, we will simulate and tune our routing extension, origination policies, wavelength-assignment and routing heuristics for optical burst switching (OBS) and optical packet switching (OPS) networks.

17 AL-FUQAHA et al.: ROUTING FRAMEWORK FOR ALL-OPTICAL DWDM METRO AND LONG-HAUL TRANSPORT NETWORKS 1459 REFERENCES [1] G. Apostolopoulos, R. Guerin, and S. Kamat, Implementation and performance measurements of QoS routing extensions to OSPF, in Proc. IEEE 18th Annu. Joint Conf. Computer Communications, Mar. 1999, pp [2] C. Assi et al., Optical networking and real-time provisioning: An integrated vision for the next-generation Internet, IEEE Network Mag., vol. 15, pp , July/Aug [3] K. Chan and T. P. Yum, Analysis of least congested path routing in WDM lightwave networks, in Proc. INFOCOM, June 1994, pp [4] R. Coltun, The OSPF opaque LSA option,, RFC 2370, [5] H. Harai et al., Performance of alternate routing methods in all-optical switching networks, in Proc. IEEE INFOCOM, 1997, pp [6] K. Kompella and Y. Rekhter, OSPF extensions in support of generalized MPLS, Internet Draft, Work in Progress, Aug [7], Routing extensions in support of generalized MPLS, Internet Draft, Work in Progress, Aug [8] K. Lee and V. O. K. Li, A wavelength-convertible optical network, J. Lightwave Technol., vol. 11, pp , May [9] B. Li, X. Chu, and K. Sohrabi, Routing and wavelength assignment versus wavelength converter placement in all-optical networks, IEEE Opt. Commun., vol. 1, pp. S22 S28, Aug [10] A. Mokhtar and M. Azizoğlu, Adaptive wavelength routing in all-optical networks, IEEE/ACM Trans. Networking, vol. 6, pp , Apr [11] S. Ramamurthy and B. Mukherjee, Fixed-alternate routing and wavelength assignment in wavelength routed optical networks, in Proc. IEEE GLOBECOM, 1998, pp [12] S. Subramaniam, M. Azizoglu, and A. K. Somani, On optimal converter placement in wavelength-routed networks, IEEE/ACM Trans. Networking, vol. 7, pp , Oct [13] G. Xiao and Y. Leung, Algorithms for allocating wavelength converters in all-optical networks, IEEE/ACM Trans. Networking, vol. 7, pp , Aug [14] H. Zang et al., A review of routing and wavelength assignment approaches for wavelength-routed optical WDM networks, Opt. Networks Mag., vol. 1, pp , Ghulam M. Chaudhry (M 85 SM 98) received the M.S. and Ph.D. degrees in computer engineering from Wayne State University, Detroit, MI, in 1985 and 1989, respectively. Currently, he is an Associated Professor in the Department of Electrical and Computer Engineering, University of Missouri Kansas City. His teaching/research interests include computer architectures, parallel and distributed systems, Verilog HDL, computer network management, and VLSI. He has published extensively in national and international journals and conferences. He has served on the steering/program committees of several international conferences and on the editorial boards of the journals. Prof. Chaudhry has served as General Chair for the International Conference on Parallel and Distributed Computing Systems in He is a member of ISCA, the Association for Computing Machinery (ACM), and IASTED. Mohsen Guizani (S 83 M 90 SM 98) received the B.S. (with distinction) and M.S. degrees in electrical engineering, and the M.S. and Ph.D. degrees in computer engineering from Syracuse University, Syracuse, NY, in 1984, 1986, 1987, and 1990, respectively. Currently, he is a Professor and the Chair of the Computer Science Department, Western Michigan University, Kalamazoo. His research interests include computer networks, wireless communications and computing, and optical networking. His research has been supported by Sprint, Telcordia, the U.S. Navy, and Boeing, to name afew. Dr. Guizani is a member of the IEEE Communications Society, the IEEE Computer Society, the American Society for Engineering Education (ASEE), the Association for Computing Machinery (ACM), the Optical Society of America (OSA), SCS, and Tau Beta Pi. Ala I. Al-Fuqaha (S 00 M 04) is a Senior Member of Technical Staff at LAMBDA Optical Systems, Reston, VA, where he works on the design and development of embedded routing protocols and network management systems for all-optical transport networks. Before joining LAMBDA, he was a Software Engineer with Sprint Telecommunications Corporation, where he worked on different projects as part of the architecture team to design and develop software to manage Sprint s core network (Nortel DMS-250/300, ATM, SS7). His research interests include high-speed computer and telecommunication networks, optical transport networks, wireless networks, network security, network management, embedded software, expert systems, distributed processing, simulation modeling, and computer architecture. Miguel A. Labrador (M 96 SM 04) received the M.S. degree in telecommunications and the Ph.D. degree in information science with concentration in telecommunications from the University of Pittsburgh, Pittsburgh, PA, in 1994 and 2000, respectively. Before joining the Department of Computer Science and Engineering, University of South Florida, Tampa, as an Assistant Professor in 2001, he was with Telcordia Technologies, Inc., as a Consultant in the Broadband Networking Group of the Professional Services Business Unit. He is currently on the Editorial Board of Computer Communications. His research interests are in the areas of transport layer protocols, active queue management, and optical networking. Dr. Labrador has served as Technical Program Committee Member of many IEEE conferences, and was the former Secretary of the IEEE Technical Committee on Computer Communications.