fault recovery in mpls network using case-based reasoning

the current internet inherently has a degree of survivability due to the connection less ip protocol. dynamic routing protocols are designed to react to faults by changing routes when routers learn about topology changes via routing information updates (e.g., link status advertisements). loss of quality of service (qos) has not been an issue because current internet traffic is the best-effort. on the contrary, the multi protocol label switching (mpls) approach is connection-oriented, which implies greater potential vulnerability to faults. at the same time, mpls will support integrated services, which are more sensitive to loss of service. reliability is becoming more important as more users depend on the internet for critical communication services and expect a higher level of performance. usually, fault recovery is attempted first at the lowest layer, and escalated to the next layer if recovery was unsuccessful or impossible. fault recovery capabilities in the mpls layer are needed as well to decouple mpls from dependence on physical layer fault recovery mechanisms which may differ between networks. this paper proposes an enhanced-scheme for fast rerouting to pre-assigned label-switched paths (lsps) in case of lsp or link failures. in order to minimize backup resources, it allows possibility of splitting traffic of faulty lsp onto available alternative lsps for fault recovery. we use pre-assigned backup lsps for restoration, when fault occurs. total traffic throughput and resource utilization can be maximized if the traffic of faulty lsp is split over multiple pre-assigned lsps. in this paper a new approach to providing fault tolerance in mpls networks using case-based reasoning (cbr) as a method to find out the amount of traffic forwarded on each pre-assigned lsp based on past experiences of loading process is presented. the pre-assigned lsps and the percentage of traffic splitting are calculated off-line based on desired qos and capacity constraints. also we evaluate the operation of successful decomposition of traffic based on the two mentioned constraints by using cbr, when the number of backup lsps as a complexity factor increase. in another point of view, in cases when there is no possibility of using the experiences successfully, there would be no other way than using the erroneous unsuccessful experiences. we thus solved our recovery problem by using, first incorrect databases in our experiments, to moving later towards decreasing the error rate in a gradual manner.