We consider extensions of the most common mesh-restorable network capacity design formulation that enhance the dual-failure restorability of the designs. A significant finding is that while design for complete dual-failure restorability can require triple the spare capacity, dual failure restorability can be provided for a fairly large set of priority paths with little or no more spare capacity than required for single-failure restorability. As a reference case we first study the capacity needs under complete dual-failure restorability. This shows extremely high capacity penalties to support 100% dual-failure restorability. A second design model allows a user to specify a total capacity budget limit and obtain the highest average dual-failure restorability possible for that investment limit. A third design strategy supports multiple-restorability service class definitions at minimum total cost. Restorability can range from best-efforts-only on any failure to an assurance of complete single and dual-failure restorability, on a per-demand basis. This work shows how to economically support an added service class in the upward quality direction: assured dual failure survivability. This lets a network operator tailor the investment in capacity to provide ultra-high availability on a selective basis, while avoiding the very high investment required for complete dual-failure restorability for all.

Most research to date in survivable optical network design and operation, focused on the failure of a single component such as a link or a node. A double-link failure model in which any two links in the network may fail in an arbitrary order was proposed recently in literature. Three loop-back methods of recovering from double-link failures were also presented. The basic idea behind these methods is to pre-compute two backup paths for each link on the primary paths and reserve resources on these paths. Compared to protection methods for single-link failure model, the protection methods for double-link failure model require much more spare capacity. Reserving dedicated resources on every backup path at the time of establishing primary path itself would consume excessive resources. In Ref. 2 and 3, we captured the various operational phases in survivable WDM networks as a single integer programming based (ILP) optimization problem. In this work, we extend our optimization framework to include double-link failures. We use the double-link failure recovery methods available in literature, employ backup multiplexing schemes to optimize capacity utilization, and provide 100\% protection guarantee for double-link failure recovery. We develop rules to identify scenarios when capacity sharing among interacting demand sets is possible. Our results indicate that for the double-link failure recovery methods, the shared-link protection scheme provides 10-15\% savings in capacity utilization over the dedicated link protection scheme which reserves dedicated capacity on two backup paths for each link. We provide a way of adapting the heuristic based double-link failure recovery methods into a mathematical framework, and use techniques to improve wavelength utilization for optimal capacity usage.

We study the total capacity requirements of span-restorable mesh network designs as the percentage of all possible dual failure combinations incident on a common node is increased. Our interest is in questions such as: Are there any guidelines or insights as to how many such SRLGs can be sustained before the capacity penalty becomes severe? Can we diagnose which SRLGs are the most limiting to overall network efficiency? When would it be worthwhile to take physical measures to eliminate a certain SRLG? In addressing these questions we provide a design formulation and procedure for planning any span-restorable network for a known set of SRLGs. One finding of interest is that if all dual failure combinations incident to a common node are allowed for in the design, then nearly all other dual span failure combinations (any two spans in the network) will also be restorable. We also produce experimental results showing how total capacity depends on the relative number or frequency of co-incident SRLGs and quantify how the type of SRLG will impact design costs.

In this paper, we study restoration mechanisms to handle channel and link failures in an optical WDM wavelength-routed wide-area backbone network based on a mesh topology. The solution uses a small number of tunable lasers per link to provide restoration capability. In addition to link failures, we consider individual channel failures link-level mechanisms are presented: redirection algorithm (RDA) and disjoint path algorithm (DPA). These mechanisms use WDM-specific link information to compute the link restoration routes. We present results based on discrete-event simulations to understand the performance of the proposed mechanisms, in terms of restoration efficiency and restoration times. The results show that for networks of varying size and node degree with 32 wavelengths on each link, using as few as 8 tunable lasers per link provides good restoration efficiency under moderate traffic load. which might occur when one or more transceivers fail at the source of the lightpath or due to a failure in the switch fabric. Restoration is first attempted using the tunable lasers to transmit on the failed wavelengths. If all the failed lightpaths cannot be restored using the tunable lasers, unused wavelengths on the same link are used (this requires optical wavelength conversion at the nodes). For the remaining lightpaths requiring restoration, link-level restoration mechanisms are attempted. Two different

In this paper, we have proposed an integrated lightpath traffic transport approach to groom and transport the client traffic onto a dynamic logical topology established over the optical network. The integrated approach has been studied against the single hop and multi-hop approaches to demonstrate its benefit on improving the performance. We have implemented heuristics for the single hop, multi-hop approaches, and three variants of the integrated approach. Three traffic patterns and two types of bandwidth distributions, totaling six traffic scenarios, have been studied with regard to the blocking probability and throughput.

In this paper we study an approach to Quality of Service that offers end-users the choice between two classes of service defined according to their level of transmission protection. The first class of service, called Fully Protected (FP), offers end-users a guarantee of survivability in the case of a single failure; all FP traffic is protected using either a 1:1 or 1+1 protection scheme at the WDM layer. The second class of service, called Best-Effort Protected (BEP), is not protected; when a failure occurs, the network does the best it can by restoring at the IP layer only as much BEP traffic as possible. The FP service class mimics what Internet users receive today. The motivation of this approach is to increase the amount of bandwidth used on backbone networks by offering a lower quality of service that does not affect the current QoS provided by the network. We design an ILP model, for finding primary and backup paths at the optical layer, that incorporates a number of carriers' common practices. Namely we allow the FP demand to be specified via a traffic matrix at the IP layer, we include an overprovisioning factor that specifies the portion of each link that must be left unused, and we incorporate a minimal fairness requirement on how the BEP traffic is allocated. Our goal is thus to quantify how much BEP traffic can be carried in addition to the FP traffic, without impacting the protection quality of the FP traffic even in the case of failure, and without impacting the FP load. We show that by having two such classes of service, the load on a network can be increased by a factor of 4 to 7 (depending upon the network). Even if carriers want to overprovision their networks by 50%, we can still triple the total network load. We illustrate that the location of the bottleneck can affect whether or not we see a difference in performance between 1:1 or 1+1 protection schemes. Finally we evaluate the tradeoff between the two carrier requirements of overprovisioning and minimal f

To accommodate the exponential growth of Internet traffic, wavelength division multiplexing (WDM) optical networks have been widely deployed in modern communication networks. IP (internet protocol) over WDM is being envisioned as a more efficient solution compared to other technologies (e.g., IP over ATM over SONET/SDH over WDM, IP/SONET/WDM, etc.). In this paper, we discuss current issues and challenges in traffic engineering for IP over WDM networks. In particular, we investigate practical aspects such as architectural alternatives, routing, signaling and survivability. Then a constraint-based routing scheme in the GMPLS context is proposed to meet different optical quality of service requirements. The performance of the algorithm is studied using simulations.

We consider optical networks with nodes interconnected by links comprising multiple channels. The blocking performance of such networks depends on the channel-switching capabilities of the nodes. In this paper, we focus on developing analytical models for evaluating the blocking performance in circuit-switched optical networks. Several architectures for channel-switching are presented. Multi-rate circuits may be established if different circuits are allocated different number of channels. Depending on how the network can assign the channels to circuits that request more than one channel, it is classified as a Data Splitting Network (DSN) or a non-DSN. We consider multi-rate circuits which require either a single channel or a given number of channels k (>1). Analytical models for computing blocking probabilities are developed for various channel-switching architectures at the nodes. The validity of the models is shown by comparing the analytical results with simulations. Numerical results in a single-fiber TDM wavelength-routing network suggest that limited channel-switching may be sufficient even for circuits that require more than one channel or slot. A comparison of DSN and non-DSN shows that data splitting can accommodate more multi-slot circuits at the expense of blocking more single-slot circuits.

In Optical Burst Switching (OBS), data packets at the edge of the network are aggregated into larger units identified as Data Bursts (DBs). When switching is performed synchronously in OBS core nodes (slotted switching), each DB has to be segmented at the switch input into fixed size units. Each unit is switched to a designated output in an "optical" time slot (a parameter of the "optical" switch). The purpose of our study is to give recommendations concerning the optimum size of these optical slots so as to minimize the overhead arising from the segmentation process. In order to estimate the total overhead due to this process we take into account the statistical distribution of the burst size, the possible padding of the last burst segment (to completely fill an optical slot) and the overhead due to the optical-slot preamble.

Optical burst switching (OBS) has recently attracted increasing interest as a network architecture for the future optical Internet. In OBS, efficient contention resolution is a key issue. This paper investigates the performance of shared converter pools for contention resolution in OBS. First, key design parameters for contention resolution in OBS nodes which employ wavelength converters and simple fiber delay line (FDL) buffers are discussed. Then, the performance of a converter pool is evaluated for a Poisson and a self-similar traffic model. Depending on load and node dimensioning, the number of wavelength converters can be reduced by 50-75%. Finally, different converter usage strategies for the combination of a converter pool and an FDL buffer are presented and compared. A strategy that prefers FDL's over converters for contention resolution can reduce burst loss probability for converter pools with a small number of converters.

A mechanism that is often proposed for Quality-of-Service differentiation in Optical Burst-Switched networks is offset-time management. In this paper we identify and explain some undesired characteristics of this mechanism. The most important finding is that the burst drop probability differentiation that is attained for a given offset-time value strongly depends on the distribution of the burst durations. Hence control of the differentiation is difficult, since the distribution of burst durations is subject to changes all the time, depending on the traffic conditions at the edge of the OBS network. We also found that offset-time management slightly increases the unfairness within the lower priority classes, in the sense that longer bursts are dropped more frequently than shorter ones.

In this paper, we propose a threshold-based burst assembly scheme in conjunction with a burst segmentation policy to provide QoS in optical burst switched (OBS) networks. Bursts are assembled at the network edge by collecting packets that have the same QoS requirements. Once the number of packets in a burst reaches a threshold value, the burst is sent into the network. We investigate various burst assembly strategies which differentiate bursts by utilizing different threshold values or assigning different burst priorities to bursts that contain packets with differing QoS requirements. The primary objective of this work is to find the optimal threshold values for varous classes of bursts. We show through simulation that there is an optimal value of burst threshold that minimizes packet loss for given network parameters.

We present an accurate model for burst traffic characteristics inside optical burst switched networks. At the edges of such networks, a number of IP packets are collected before being injected into the core as a single unit. The sizes of these so-called bursts critically impact the performance of the various network elements, so that a good understanding of their characteristics is an essential step in network performance engineering. A further abstraction of this assembly process results in a straightforward simulation model that can eliminate the need for packet level simulation, and thus reduce model complexity and simulation cost.

Optical Burst Switching (OBS) is considered as a promising switching technique for building the next generation optical Internet. In OBS networks, one important issue is how the performance will be affected by bursts assembled from packets, which is the basic transmission unit in OBS. In this paper, we study the fundamental statistic properties such as the burst length distribution, inter-arrival time distribution, as well as correlation structure of assembled burst traffic from burst assembly algorithms. From both theoretical and empirical results, it is demonstrated that after the assembly, the traffic will in general approach the Gaussian distribution. In particular, the variance of assembled traffic decreases with the increase in the assembly window size and the traffic load. However, the long range dependence in the input traffic will not change after assembly. Such smoothed assembled traffic will enhance the OBS performance by reducing burst loss and increase OBS throughput. This result is useful for the future study of OBS node and networks.

We present a formal protocol description for a Just-In-Time (JIT) signaling scheme running over a core dWDM network which utilizes Optical Burst Switches (OBS). We apply an eight-tuple extended finite state machine (EFSM) model to formally specify the protocol. Using the EFSM model, we define the communication between a source client node and a destination client node through an ingress and one or multiple intermediate switches. We worked on single burst connections that means setting up the connection just before sending a single burst and then closing the connection as soon as the burst is sent. The communication between the EFSMs is handled through message transfer between protocol entities.

Signaling protocols in switches are primarily implemented in software for two important reasons. First, signaling protocols are quite complex with many messages, parameters and procedures. Second, signaling protocols are updated frequently requiring a certain amount of flexibility for upgrading field implementations. While these are two good reasons for implementing signaling protocols in software, there is an associated performance penalty. Even with state-of-the-art processors, software implementations of signaling protocol are rarely capable of handling over 1000 calls/sec. Correspondingly, call setup delays per switch are in the order of milliseconds. Towards improving performance we implemented a signaling protocol in reconfigurable FPGA hardware. Our implementation demonstrates the feasibility of 100x-1000x speedup vis-a-vis software implementations on state-of-the-art processors. The impact of this work can be quite far-reaching by allowing connection-oriented networks to support a variety of new applications, even those with short call holding times.

Sparse wavelength conversion and appropriate routing and wavelength assignment (RWA) algorithms are the two key factors in improving the blocking performance in wavelength-routed all-optical networks. It has been shown that the optimal placement of a limited number of wavelength converters in an arbitrary mesh network is an NP complete problem. There have been various heuristic algorithms proposed in the literature, in which most of them assume that a static routing and random wavelength assignment RWA algorithm is employed. However, the existing work shows that fixed-alternate routing and dynamic routing RWA algorithms can achieve much better blocking performance. Our study in this paper further demonstrates that the wavelength converter placement and RWA algorithms are closely related in the sense that a well designed wavelength converter placement mechanism for a particular RWA algorithm might not work well with a different RWA algorithm. Therefore, the wavelength converter placement and the RWA have to be considered jointly. The objective of this paper is to investigate the wavelength converter placement problem under fixed-alternate routing algorithm and least-loaded routing algorithm. Under the fixed-alternate routing algorithm, we propose a heuristic algorithm called Minimum Blocking Probability First (MBPF) algorithm for wavelength converter placement. Under the least-loaded routing algorithm, we propose a heuristic converter placement algorithm called Weighted Maximum Segment Length (WMSL) algorithm. The objective of the converter placement algorithm is to minimize the overall blocking probability. Extensive simulation studies have been carried out over three typical mesh networks, including the 14-node NSFNET, 19-node EON and 38-node CTNET. We observe that the proposed algorithms not only outperform existing wavelength converter placement algorithms by a large margin, but they also can achieve almost the same performance comparing with full wavelength conversion under the same RWA algorithm.

In this paper, we address the problem of wavelength band switching (WBS) in multi-granular, wavelength division multiplexing optical networks. The purpose of WBS is to reduce the increasing control complexity and port numbers in ordinary optical cross-connects. We classify the WBS problem into eight variations and then focus on one of them, wherein there are a fixed number of wavebands in each fiber, each with a fixed number of wavelengths and in fact, consisting of a fixed set of wavelengths. We develop an integer linear programming (ILP) model, which for a given set of lightpath requests, determines the routes and assigns wavelengths for the lightpaths so as to minimize the number of ports needed. Our model is more generic than existing solutions, as it takes into account the grouping of lightpaths having different sources and destinations and hence maximizes the benefits of WBS. Since the subproblem of minimizing the port count with WBS is NP-complete, we also develop and compare two heuristic algorithms: namely Waveband Oblivious optimal Routing and Wavelength ssignment (or WBO-RWA), and Balanced Path routing with Heavy-Traffic first waveband assignment (or BPHT). For small networks, our results indicate that the performance of the BPHT heuristic is quite close to that achievable using the ILP. For larger networks, the ILP model is no longer feasible, but BPHT can perform onsiderably better than WBO-RWA. Our results also provide valuable insights into the effect of wavelength band granularity on the performance of WBS algorithms, as well as the tradeoffs between the wavelength-hop and the port count.

Wavelength division multiplexed (WDM) networks are matured to provide, scalable data centric infrastructure, capable of delivering flexible, value added, high speed and high bandwidth services directly from the optical (WDM) layer. But, providing fault-tolerance at an acceptable level of overhead in these networks has become a critical problem. Several methods exist in the literature which attempt to guarantee recovery in a timely and resource efficient manner. These methods are centered around a-priori reservation of network resources called spare resources along a node disjoint protection path. This protection path is usually routed from source to destination along a totally link disjoint path from primary path. This paper considers the problem of routing and wavelength assignment (RWA) in wavelength routed WDM optical networks. In particular, we propose an efficient algorithm to select routes and wavelengths to establish dependable connections (D-connections), called segmented protection paths. Our algorithm does not insist on the existence of totally disjoint paths to provide full protection. We present experimental results which suggest that our scheme is attractive enough in terms of average call acceptanc ratio, spare wavelength utilisation, and number of requests that can be satisfied for a given number of wavelengths assuming that the requests come one at time, and wavelengths are assigned according to fixed ordering.

In this paper, we have developed an integrated online algorithm for dynamically routing bandwidth guaranteed label switched paths (LSPs) in IP over WDM optical networks. Traditionally, routing at upper layer (e.g., IP layer) is independent of wavelength routing at the optical layer. Wavelength routing at the optical layer sets up a quasi-static logical topology which is then used at the IP layer for IP routing. The coarse-grain wavelength channels and the pre-determined virtual topologies with respect to some {\it a priori} assumed traffic distribution are barriers to efficient resource use and inflexible to changing traffic. We take into count the combined knowledge of resource and topology information in both IP and optical layers. With this added knowledge, an integrated routing approach can extract better network efficiencies, be more robust to changing traffic patterns at the IP layer than schemes that either use dynamic routing information at the IP layer or use a static wavelength topology only. Our proposed algorithm considers not only the importance of critical links, but also their relative importance to routing possible future LSP set-up requests by characterizing their normalized bandwidth contribution to routing future LSP requests. Moreover, link residual bandwidth information that captures the link's capability of routing future LSPs is also incorporated into our algorithm. Simulation results show that our proposed algorithm performs better than both routing algorithms do in terms of LSP set-up request rejection rate and available bandwidths.

The effects of an attack connection can propagate quickly to different parts of a transparent All-Optical Network. Such attacks affect the normal traffic and can either cause service degradation or outright service denial. Quick detection and localization of an attack source can avoid losing large amounts of data in an All-Optical Network. Attack monitors can collect the information from connections and nodes for diagnostic purpose. However, to detect attack sources, it is not necessary to put monitors on all nodes. Since those connections affected by the attack connection would provide valuable information for diagnosis, we show that placing a relatively small number of monitors on a selected set of nodes in a network is sufficient to achieve the required level of performance. However, the monitor placement, routing, and attack diagnosis are challenging problems which need research attention. We, in this paper, first develop our models of crosstalk attack and monitor node. With these models, we prove the necessary and sufficient condition for one-crosstalk-attack diagnosable network. After that, we develop a scalable diagnosis method which can localize the attack connection efficiently with sparse monitor nodes.

Gain competition in an optical amplifier can result in a performance-degrading reduction in the gain of a channel if the overall input power of the amplifier is increased. A gain competition attack may be realized by one or more attackers (pretending to be legitimate users) increasing their source powers in order to degrade the quality of service seen by other users. In this paper, we study the effect of an optical amplifier gain competition attack in a point-to-point WDM link. By looking at the relative OSNR degradation ratio of the attacked channels, we show that Automatic Gain Control mechanisms can alleviate the absolute OSNR degradation to a significant extent, but cannot immunize the system from performance deterioration if the attacking user's power is strong enough. Adding more amplifiers to a link will enhance the service quality as well as the system's robustness against a gain competition attack at the price of higher network cost.

In this paper, we comparatively evaluate two photonic packet switch architectures with WDM-FDL buffers for synchronized variable length packets. The first one is an output buffer type switch, which stores packets in the FDL buffer attached to each output port. Another is a shared buffer type switch, which stores packets in the shared FDL buffer. The performance of a switch is greatly influenced by its architecture and the packet scheduling algorithm. We compare the performance of these two packet switches by applying different packet scheduling algorithms. Through simulation experiments, we show that each architecture has a parameter region for achieving a better performance. For the shared buffer type switch, we found that void space introduces unacceptable performance degradation when the traffic load is high. Accordingly, we propose a void space reduction method. Our simulation results show that our proposed method enables to the shared buffer type switch to outperform the output buffer type switch even under high traffic load conditions.

This paper presents the performance analysis of an optical packet switch with a feedback delay line buffer, performed in the framework of the IST-DAVID research project. The network operation is slotted and the case of variable length data bursts fitted into trains of fixed size packets is considered. The role of time and wavelength domains for congestion resolution is investigated, showing that they must be exploited as combined tools in order to obtain good node performance and, at the same time, optimize the optical hardware requirements.

Survivability to faulty components and simplified management drive the practical deployment of ring-based WDM networks. In many applications, location constraints and user scalability require that multiple rings are interconnected to form a single large network. Survivability of connections spanning across multiple rings is then achieved by resorting to dual-interconnection, i.e., two (or more) nodes are available to crossconnect the inter-ring traffic between two neighboring rings. By providing one backup crossconnect-node to be used in case of failure of the primary crossconnect-node, network wide connectivity is thus guaranteed also in presence of any faulty node. This paper addresses the problem of optimally provisioning both bandwidth and crossconnect ports required to satisfy a set of traffic demands in a dual-interconnected WDM dual-ring network architecture. The problem is solved under two design scenarios. In the first scenario, priority is given to the minimization of the number of wavelengths. In the second scenario, priority is given to the balancing of traffic between the crossconnect-nodes. Two efficient approaches are proposed that provide a near-optimal solution in each considered scenario. The discussed performance comparison provides the network designer with a quantitative assessment of the trade-off between the two approaches.

We investigated the photonic packet switch architecture that enables a high node throughput and provides priority services. Previously, we proposed PBSO (partial buffer sharing with overwriting) method that enables prioritized buffer management to control an optical buffer. The PBSO method is based on a single queue and its complexity is O(p), where p is the number of priority classes. PBSO can be used to provide different levels of drop precedence in DiffServ (Differentiated Services) Assured Forwarding. In this paper, we propose an analytical method of PBSO where p=2. We assess the accuracy of the analytic method. We show that PBSO improves the packet loss probability in each priority class more than the existing PBS (partial buffer sharing) does, and that it can be used for prioritized buffer management of an optical buffer. PBSO is especially effective when the arrival rate of higher priority class packets is much lower than that of lower priority class packets. In this case, PBSO dramatically improves the performance of higher priority class packets while the degradation in the performance of lower priority class packets is small. In PBSO, a larger number of higher priority class packets can be accepted at a given packet loss probability than in PBS or non-priority method.

This paper solves the problem of path selection for WDM mesh networks with a special focus on the implementation in middle-sized networks, such as metropolitan-area networks (MANs). A novel routing and signaling protocol, called Asynchronous Criticality Avoidance (ACA), is proposed to improve the network performance. With the ACA protocol, a specific set of wavelength channels are defined as critical links between a node pair according to dynamic link-state. Criticality information is defined as the critical links and the associated information, which is coordinated and disseminated by each source node to every other source node as an inter-arrival planning. Routing and wavelength assignment is performed along with the criticality avoidance mechanism, in which path selection process is devised to take the criticality information into consideration. Simulation is conducted in 22- and 30-node networks to examine the proposed approach. The simulation results show that the ACA protocol significantly outperforms the Fixed-Path Least-Congested (FPLC) scheme under the Fixed Alternate Routing architecture with various patterns of alternate paths assigned to each source-destination pair in the networks.