| Summary: | The rapid growth of the data traffic, caused by the ever-growing number of Internet users and the services requiring huge
bandwidth (e.g., videoconferences, IPTV, etc.) show that the existing transport network infrastructure is not optimal in terms of
cost, flexibility and scalability. In this context, an appealing solution relies on deploying a multi-layer network (MLN) constituted
by a dual layer infrastructure combining the benefits provided by flexible Connection Oriented (CO) - Ethernet and the huge
transmission capacity provided by Wavelength Switched Optical Networks (WSON), controlled by a Generalized Multi-Protocol
Label Switching (GMPLS) unified control plane.
An important concept in such a MLN is the Forwarding Adjacency (FA). The aim of FAs is to exploit the grooming decisions
where existent lower-layer Label Switched Paths (LSPs) (i.e., FA LSPs) with sufficient unreserved bandwidth are intelligently
reused when accommodating and serving new upper-layer LSPs. By doing so, the established lower-layer LSPs form a
Virtual Network Topology (VNT) for the provisioning of CO-Ethernet services.
One of the main challenges in a MLN is how to accommodate dynamically requested higher-layer connections in the created
VNT to optimally utilize the large capacity offered by a WSON. In this regard, we focus on the problem of path computation and
LSP provisioning in a CO-Ethernet over WSON network, under the framework of GMPLS unified control plane. Firstly, we
propose the usage of a FA TE link timer, which postpones the release of created FA TE link when there are no more upperlayer
connections over such a link, rather than releasing it immediately. By doing so, we do reduce the signaling overhead
due to consecutive triggering of FA TE links between the same pair of nodes. As a second contribution, we compare three
approaches for dynamic VNT reconfiguration: semi-dynamic, virtual and dynamic, using our proposed online path
computation algorithm. We show that with the dynamic approach the lowest connection blocking probability is obtained but at
the expense of the increased setup delay. Finally, we have proposed a dynamic path computation algorithm that chooses the
route depending on the current state of the network resources, favoring the usage/re-use of virtual and active FA TE links over
establishing new optical LSPs occupying unused wavelength channels. The proposed algorithm provides better usage of the
network resources, keeping the connection blocking probability low, comparing with selected algorithms proposed in the
literature.
Next, we focus on schemes and mechanisms for dedicated path protection in a CO-Ethernet over WSON network. In that
sense, after classifying recovery mechanisms for MLN found in the literature, we compare both link- and Shared Risk Link
Group (SRLG) - disjoint schemes. These schemes aim at computing a link- and SRLG-disjoint backup path with respect to
the computed working paths, respectively. It is shown that, although the SRLG-disjoint scheme provides higher survivability
comparing with the link-disjoint scheme, the connection blocking probability is still significantly higher. As a second
contribution regarding dedicated path protection, we present a protection scheme which tries to accommodate a requested
connection over the least congested TE links (i.e., higher unused bandwidth), while establishing both, working and backup
paths. The aim of this contribution is to provide an efficient usage of the network resources and to minimize the number of
connections affected by a link failure. Finally, we propose a Suurballe algorithm-based scheme that aims at decreasing the
connection blocking probability due to working path SRLG-joint links that are removed during the backup path computation.
The exhaustive performance evaluations of the above contributions are conducted through simulations under the assumption
of dynamic traffic pattern and using well-known reference network topologies.
|
|---|
