| Summary: | 博士 === 國立中正大學 === 電機工程所 === 98 === Network survivability is one of the most critical issues. With the growing of Internet, a failure would result in a lot of packet losses. In order to mitigate the risk, many approaches were presented to design a survivable network. In this thesis, we focus on three kinds of network survivability problems, (1) Shortest path tree (SPT) restoration problem, (2) IP fast reroute problem and (3) Load balanced IP protection problem.
For SPT Restoration problem, we used network simplex method (NSM) to compute SPT and adopted sensitivity analysis (SA) technique to analyze the relation between SPT and link metric (or cost). Through the SA technique, once the metric of a link changes, the new SPT can be computed with low complexity.
For IP fast reroute problem, we proposed a high survivability reroute scheme to handle single link/node failure for IP networks, called Priority Loop-free Alternative Neighbors (PLAN) scheme. The main idea is to differentiate the routing capability of adjacent neighbors and select them in a priority order to be used as alternatives for forwarding packets when a link/node failure occurs. Besides, PLAN avoided the occurrence of a routing loop at anytime during the healing process, while achieving high survivability.
As for IP protection problem, we proposed two load balanced fast local IP protection schemes, we call Simulated Annealing based Load Balanced (SALB) and Non-Weighted Interface Specific Routing (NISR). SALB adopts the concept of simulated annealing to tune link metric so that the working traffic can be routed as balanced as possible and guaranteed against any single link failure. By SALB scheme, the working paths are routed along with shortest paths and backup paths are pre-computed and stored in the backup routing tables. Once a failure occurs, the routers next to failure use the pre-built backup routing tables to rapidly divert the packets carried on failed component.
NISR scheme determines the working and backup routing tables of IP routers before any single link failure. It jointly considers protection switching time, network survivability, and traffic load distribution together. Unlike conventional IP routing, in NISR, we relax shortest path routing in computing working and backup routing tables. Most importantly, each interface in a router has its own routing tables. We formulate this problem as a mixed integer linear programming problem in which the traffic load on the most congested link is to be minimized. Since this problem is intractable by its NP-hard nature, we further decompose it into several sub-problems. Each sub-problem is solved optimally and the solutions of those sub-problems are combined to provide a solution to the original problem.
We carry out our experiments on benchmark networks and compare all proposed schemes to other existing approaches for the three problems, respectively. Numerical results delineate that these proposed schemes can significantly enhance network survivability.
|
|---|
