IP Fast Reroute (IPFRR) is a technique aiming to reduce packet loss during routing convergence. With IPFRR, a router pre-computes protection paths which bypass certain links/nodes on its default shortest path. A router can quickly switch to a protection path once it locally detects the failure, which does not rely on routing convergence. NotVia is an IPFRR technique currently under standardization, and is also the only technique which can protect all single link and node failures. However, the computation and storage of protection paths introduce non-trivial overheads, which make network operators hesitate to deploy it. Furthermore, NotVia is not management friendly, as each router does not know what protection paths it is on. The information is critical for anticipating re-routing traffic so that extra bandwidth could be provisioned beforehand.
In this project, we improve NotVia by reducing its overhead and strengthening its manageability. We identify two useful techniques to reduce the overhead: Aggregation and Prioritization. In terms of manageability, we develop an algorithm called rNotVia which can efficiently compute the necessary management information. The techniques are evaluated using a customized simulator based on real ISP topologies, and the results suggest significant saving of both computational load and memory consumption could be achieved.
The customized simulator we used to generate the results in our CoNext'07 paper.
Any feedback, suggestions, and questions could be sent to angl AT cs.duke.edu