Recent advances in datacenter network design have enabled full bandwidth communication based on the notion of Valiant load balancing. In this paper, we study from the link capacity viewpoint how such communication paradigm can be supported in the context of link failures, the study of which is currently absent. In particular, we target full bandwidth communication among all the servers, for all valid traffic patterns, and under k arbitrary link failures. We derive the minimum link capacity required on two typical datacenter topologies—VL2 and fat-tree. Our main finding is that given the same server scale, fat-tree requires less total link capacity than VL2 for 1 ≤ k ≤ n over 4, where n denotes the port count of homogeneous switches used in fat-tree. For k > n over 4, there exists a turning point from which VL2 becomes more capacity-efficient.

This paper proposes a new monitoring path design paradigm for independent dual link failures. Specifically, the new approach exploits the sequential arrival and departure property of independent failure events to uniquely localize failed links. Such property, however, cannot be captured by the existing approach, which is built upon the notion of shared risk link groups. Consequently, we show via solution space comparison that the existing approach can result in overdesign in terms of monitoring resources required. Numerical results further indicate that the new approach outperforms the existing one in terms of monitoring cost and computational efficiency.

We propose an analytical model that computes availability of upper-layer flows in two-layer networks with dedicated path protection deployed at the upper layer. Our proposed model reduces overestimation of the existing model caused by the ignorance of upper-layer failure correlations. We compute the flow availability by properly taking into account such correlation, and tracing upper-layer failures to lower-layer root causes. Our simulation results show that our proposed model reduces overestimation on upper-layer flow unavailability by about 30%, and thus significantly relaxes unnecessary high-availability requirements on lower-layer links without compromising the availability of upper-layer flows.

We study the benefits and trade-off of using scheduled lightpaths for traffic adaption. We propose a network planning model, which allows lightpaths to slide within its desired timing window with no penalty on the optimization objective, and to slide beyond its desired timing window with a decreasing tolerance level. Our model quantitatively measures the timing satisfaction or violation. We apply the Lagrangian Relaxation and Subgradient Method to the formulated optimization problem. Our method demonstrated great computational efficiency when compared with other existing algorithms. Our simulation results show how timing flexibility improves network resource utilization and reduces rejections.

In this comment, several errors in Chlamtac et al., 1996, Lightpath (Wavelength) Routing in Large WDM Networks and one error in Xu et al., 2000, Dynamic Routing and Assignment of Wavelength Algorithms in Multifiber Wavelength Division Multiplexing Networks are pointed out. We present corrections to the errors.

We present a study on the static resource allocation in lightpath routed WDM networks, where each request is associated with a service grade. The goal is to maintain certain acceptance ratios for the requests of all grades, as well as to minimize the resource consumption. We propose a model of static grade-of-service (GoS) differentiation as minimizing the total rejection and cost penalty. Then, we use the Lagrangian relaxation and subgradient methods to solve the problem. The results of using static GoS differentiation are presented.

Various static resource allocation algorithms have been used in WDM networks to allocate resources such as wavelength channels, transmitters, receivers, and wavelength converters to a given set of static lightpath demands, based on certain design objectives. However, although optimized resource allocations can be obtained, it remains an open issue how to determine which resources are bottlenecks to achieve better performance. Existing static resource allocation algorithms do not explicitly measure the impact of a given resource on the design objective. In this paper, we propose such a measurement based on the Lagrangian relaxation (LR) framework We use the optimized values of Lagrange multipliers as a direct measurement of the criticality of resources. Such a quantitative measurement can be naturally acquired along with the optimization process to obtain the optimal solution (or a near optimal solution) to the static routing and wavelength assignment (RWA) problem. Such a measurement helps to identify critical resources, and thus to decide the best way to add or reallocate resources.

In this paper, we study the routing and wavelength assignment (RWA) problem in a semi-dynamic scenario where rearrangements are conducted in a series of sessions after traffic demands vary. Unlike pure static RWA problems, each rearrangement scheme must consider established lightpaths in the previous session. A novel formulation of the WDM network rearrangement problem is used to minimize rejected new demands and rerouted lightpaths. This is done by coordinating the re-routing of existing lightpaths with the adaptation to varying demands. The Lagrangean Relaxation and Subgradient Method (LRSM) has been successfully used to solve the problem along with fairness consideration. The superior performance and reduced computation complexity of our algorithm are demonstrated in sample networks. In addition, we evaluate the benefit of using wavelength converters in a WDM network rearrangement. In contrast to previous studies with conclusions that wavelength converters are of little value in the static RWA problem, we show that wavelength converters improve network performance in a WDM network rearrangement.

We propose a novel link-based formulation of the WDM network reconfiguration problem for the semi-dynamic scenario, where reconfigurations are required in a series of sessions as traffic demands vary. Each reconfiguration session must consider existing connections obtained from the previous reconfiguration session. We coordinate the reconfiguration of existing connections with the new traffic demands by minimizing the rejected demands and reconfigured connections. With a much more complicated formulation, we are also able to incorporate the fairness consideration. The Lagrangean Relaxation and Subgradient Methods (LRSM) are successfully used to solve the WDM network reconfiguration problem. We demonstrate superior algorithm performance and reduced computation complexity in sample networks. In addition, we demonstrate that there is a benefit of using wavelength converters in WDM network reconfiguration problem, in contrast to the conclusion of the previous studies that wavelength converters are of little value in the static RWA problem.

A network management system software is prototyped using a service-oriented architecture with its core built using Jini and JavaSpaces technologies. The management system allows users to provision connections across multiple management domains. The system's functional modules are distributed to different management domains. Since the new management system is a distributed system, it imposes special requirements on system reliability. System reliability is improved by using lease-based resource and service sharing, transaction-based resource, concurrent access and persistent storage of resource information, and federation clustering. With these system reliability enhancements, the system is able to maintain its integrity even in the event of partial system failures.

This paper proposes a scalable design for next-generation optical cross-connects (OXCs), where a novel strategy for dimensioning the network switching capability as a long term planning is presented. When traffic demand grows, the proposed scheme simply expands the switching capacity of OXCs in waveband- and fiber-switching tiers. To minimize the number of extra fibers for waveband- and fiber-switching tiers required to satisfy a given traffic matrix, we formulate the problem of routing and wavelength assignment with tunnel allocation (RWAT) into two integer linear programming (ILP) processes that are performed sequentially. Experiments are conducted on two sample networks to compare the throughput and the number of switch points when the networks adopt different switching architectures with different traffic loads. We conclude that the proposed optimization scheme can dimension the networks with expandability and scalability to the growing traffic demand.

When the control and data planes are physically separate, it is important to maintain the established connections in the data plane even if the protocol plane fails. Channel status is maintained by a signalling protocol. Channel status information will be permanently lost as a result of a failure in the control plane having no recovery capability of the constrained-based routing label distribution protocol (CR-LDP). A recovery mechanism for channel status information is proposed in this paper. A downstream node stores a label information database (LID) of assignable (free) labels for each incoming link. A copy of the LID redundantly stored in the upstream node as a label information mirror (LIM). Each LIM and the corresponding LID are synchronised. The recovery of channel status information is guaranteed.

This paper proposes a scalable design for next-generation optical cross-connects (OXCs). We present a novel strategy for dimensioning the switching capability as a long term planning. Switching fabrics in OXCs have to be expanded according to traffic growth, which may incur a scalability problem due to the exponentially increasing cost in manufacturing and maintenance. The proposed scheme expands the switching capacity of OXCs with waveband- and fiber-switching components (or, equivalently, expands the network capacity with waveband- and fiber-switching tiers). To minimize the number of extra fibers for waveband- and fiber-switching tiers required to satisfy a given traffic matrix, we formulate the problem of routing and wavelength assignment (RWA) with tunnel allocation (RWAT) into a constraint programming (CP) process. The CP is simplified as two integer linear programming (ILP) processes that are performed sequentially. Experiments are conducted on four examples to compare the throughput and the number of switching points when different switching architectures are adopted under different traffic increase. The benefits of our approach are demonstrated. Finally, we conclude that the proposed optimization scheme can dimension the networks with expandability and scalability to the growing traffic demand.

The LDP (label distribution protocol) is used in the control plane to control an optical network. The data plane and the control plane of an optical network could be physically separate. So a failure in the control plane does not necessarily imply a data plane failure and that user communications have to be interrupted. The standard LDP, however, does not provide any mechanism to recover the knowledge stored in LDP entities about the status of the data plane after the faults are fixed. This is a reliability problem of LDP and results in the unnecessary degradation of user communications. On the other hand, in MPLS-enabled IP networks, being able to recover LDP sessions would be potentially faster and more scalable than to re-establish all affected LSPs. The proposed recovery method of LDP for the control plane failures uses label information mirrors (LIMs) in upstream downstream label switching routers (LSRs). Each LIM is a copy of the label information database (LID) in the LSR of an LDP session. We propose a systematic approach to synchronize the contents of a LIM and the corresponding LID, and show how a LIM is used to handle a control plane failure. Detailed descriptions of the recovery procedure for both control channel failures and control node failures are presented. Some significant features of the proposal are outlined.