Learning Automata-Based Scalable PCE for Load-Balancing in Multi-carrier Domain Sequences

Abstract

The prior selection of a domain sequence is a key issue for an optimal establishing of inter-domain paths. The architecture based on the path computation element (PCE) for selecting the domain sequence was proposed. Inter-domain PCEs were established that could have TE information about the link status of the multi-domain network. Confidentiality fails in the case of multi-carrier domain sequences because the inter-domain PCEs are not controlled by certain network operators. The inter-domain PCEs are also exposed to overload due to the calculation of the domain sequences. Confidentiality and scalability problems are avoided by proposing the per-domain technique based on PCE, where the PCE of the source domain has learning automata (LA) for the selection of multi-carrier domain sequences. The selection of the inter-domain path is made with a low complexity from a set of paths belonging to different multi-carrier disjoint-domain sequences. The incorporation of an LA-PCE in the source domain of the connection between domains allows decreasing the blocking probability with respect to BGP. Scenarios have been used for uniform traffic load between pairs of domains as well as links exposed to congestion.