Typically, core networks are provided with both optical and electronic physical layers. However, the interaction between the two layers is at present limited, since most of the traditional transport functionalities, such as traffic engineering, switching and restoration, are carried in the IP/MPLS layer. In the light of this, the research community has paid little attention to the potential benefits of the interaction between layers, multilayer capabilities, on attempts to improve quality of service control.
This paper shows when to move incoming label switched paths (LSPs) between layers based on a multilayer mechanism that trades off a QoS metric, such as end-to-end delay, and techno-economic aspects. Such a mechanism follows the Bayesian decision theory, and is tested with a set of representative case scenarios.
Core networks are typically equipped with both electronic and optical resources. This means that incoming traffic can be routed in either the optical or electrical domain. Essentially, electronic routing has the well-known advantages of statistical multiplexing and granularity, but is a hard-computational process for high-speed networks and it further introduces queuing delay to packets. On the other hand, data packets switched in the optical domain only experience propagation delay. However, optical resources provide a granularity which is too coarse for typical Internet streams, even if they come from the multiplex of many users.
In this IP over WDM scenario new challenges appear, since it is necessary to manage two layers, which can provide some functionalities to both of them. This is the case of routing, traffic engineering, quality of service, resilience techniques, resources optimization, etc. which could be carried out in either the IP or the WDM layer. Over the few years, a considerable effort has been dedicated to the development of automatic switched optical network (ASON) and generalized multiprotocol label switching (GMPLS). Thanks to this development, a standardized control plane has been defined, which allows a framework to propose solutions to the previous problems: traffic engineering [1], routing [2] and [3] or grooming [3] and [4].
In conclusion from previous papers in this area [4], [5], [6] and [7], it is highly desirable to efficiently combine the benefits of both optical and electronic domains to solve previously cited problems. With this aim, architectures to build multilayer-capable routers have been defined [5] and [8]. In this situation, incoming label switched paths (LSPs) traverse the multilayer-capable router, which has to decide whether to perform optical or electronic switching (Fig. 1). If an incoming LSP is routed in the electronic domain, it suffers hop-by-hop opto-electronic conversion (with subsequent delay), otherwise the router provides an optical bypass. The choice of electronic or optical switching is based upon a set of previously-defined rules in the multilayer-capable router. However, these rules are still open. The authors in [9] address the multilayer traffic engineering problem, proposing a cost model based on the link occupation. Depending on the link occupation, the router is able to decide the number of LSPs switched through each lightpath. Nevertheless, no QoS evaluation, in terms of end-to-end delay, is performed, while the paper is more focused on load balancing issues. In [1], the authors propose an ILP optimization algorithm to minimize the load in the electronic domain using cut-through lightpaths, subject to the network equipment restrictions.In this paper, we propose a techno-economic model to help routers take the decision of optical or electronic switching of their LSPs. Such an approach makes use of Bayesian decision theory, and takes into account several aspects concerning the quality of service perceived by packets, by means of queuing delay, and also techno-economic aspects such as the relative cost associated to switching LSPs in either the optical or the electronic domain. The algorithm’s computational cost is low and only have to be computed when a new LSP arrives at the router or any of the input parameters of the algorithm vary. This multilayer algorithm could be easily implemented in the control unit of the multilayer-capable router (Fig. 1).
In the light of this, the remainder of this work is organized as follows: Section 2 covers the mathematical foundations for such techno-economic analysis with a Bayesian decisor. In this section, a set of experiments and numerical examples is also provided to show how to reach an optimal decision. Section 3 studies the behavior of the bayesian decisor in a dynamic environment, with its analytical definition and experiments. Finally, Section 4 outlines a summary of the results obtained and further lines of investigation.
This paper’s main contribution is two-fold: First, it presents a novel methodology, based on the Bayesian decision theory, that helps multilayer-capable routers to take the decision of either optical or electronic switching of incoming LSPs. Such decision is made based on technical aspects such as QoS constraints and long-range dependence characteristics of the incoming traffic, nonetheless it also considers the cost differences of optical and electrical switching. This way permits high flexibility to the network operator to trade-off both economic and technical aspects.
Secondly, this paper proposes the Bayesian decision theory as the mathematical framework for dealing with the decision of optical or electronic switching of LSPs. Such mathematical framework is of low complexity, and can easily adapt to changing conditions: QoS guarantees, traffic profiles, economic aspects and network operator preferences.
Finally, this algorithm can be implemented in a per node basis by using local and independent parameters (e.g delay thresholds and optical-electronic cost) in each node. However, in further extensions of this mechanism, the local QoS parameters used in each node will be based on information regarding end-to-end delay throughout the whole network.
