Fast convergence with fast reroute in IP networks

Abstract

It has been observed that even in well managed networks, failures of links and routers are not uncommon. In order to satisfy the demand for high availability in case of a failure, fast restoration of loop-free forwarding along the optimal paths is imperative for a routing scheme. Failures can be circumvented quickly with local rerouting but packets take potentially long detours. Global recomputation of new optimal routes incurs a convergence delay and can cause forwarding loops during convergence. Attempts to avoid transient loops may also increase the convergence delay. The recently proposed SafeGuard mechanism overcomes these problems, i.e., it is always loop-free, and minimizes disruption time and convergence delay. One drawback however, is that SafeGuard needs each packet to carry multiple-byte information about the path cost. We propose an alternative approach, fast convergence with fast reroute (FCFR), that employs a fast reroute scheme such as NotVia and needs just one additional bit in the packet header. We evaluate the performance of FCFR, and show that it performs comparably to SafeGuard, with much less per-packet overhead.