Providing Protection and Restoration with Distributed Multipath Routing

Multipath routing is an interesting tool to provide a fast reaction time to protect networks from failure or congestion. Indeed, a local alternate path computation allows to faster re- route the traffic without flooding the entire network with Link State Advertisements. The restoration depends therefore on the protection guaranteed by each router. Distributed techniques allow to entrust the potential restoration to each router where it is possible. We distinguish multipath routing and fast rerouting techniques to underline the possibility to use alternate routes for load balancing or just as backup solution. In this paper we first summarize our incoming interface multipath routing technique and then analyze its capabilities in terms of protection. We evaluate several routing techniques to achieve a good coverage. Results indicate that our hop by hop routing multipath protocol performs almost as well as the best unipath rerouting technique whereas it can also be used for proportional routing. Index Terms—Fault tolerance, Multipath Routing, Fast re- routing, Coverage, Path computation. I. INTRODUCTION Protection and restoration add a new layer of reliability, integrity and availability for the network resilience on every routers they are applied to. Protection is commonly defined by pre-provisioning a backup path. This ensures a quick re- covery time because the path is computed, but not necessarily activated, before failure. When a failure occurs, a restoration protocol computes an alternate path on demand, or just selects a pre-computed alternate path. IP's Interior Gateway Protocols (IGP), such as OSPF (12) or ISIS (14) have to flood the entire network when topology changes (through Link State Advertisements, LSA) and each router needs to recompute all paths with a SPF (Shortest Path First) algorithm. However for real-time service level requirements, the reaction time might be too long to offer a sufficient quality of service. Applications such as VoIP are particularly sensitive to packet loss, therefore a fast restoration scheme is vital to avoid disruptions. A restoration time lower than 50 milliseconds became a reference (6) to guarantee the efficiency of recovery techniques. There are two simple ways to reduce the time which is necessary to compute new routes. The first consists in segmenting the network in multiple areas because there are fewer routers concerned by failures. The second one aims at minimizing the time period between two consecutive Hello messages to reduce the failure detection time. Even though today links speed and CPU power are considerable, these two techniques are generally insufficient to reach a short reaction time. Another research area is the optimization of the time used to recompute the routes which have a failed link (for example the incremental SPF given in (10)) or the optimization of the FIB (Forwarding Information Base) updating procedure. All these techniques have to be paired with pre-computed alternate paths. The reaction time depends on three elements : a) Failure detection (timers issue, SDH alarms) b) Failure notification (link state broadcasting, topological database updating) c) Re-routing (path restoration and FIB updating) In this article we focus on path provisioning protocols to accelerate the rerouting phase. We specifically study recovery based on IGP such as IP fast re-routing. First, we analyze the number of validated loop free alternate routes. Then, we weight up local recovery capabilities (on the router which detects the failure) as compared to global recovery capabilities, on routers which are located upstream from the failure. The main difficulty to achieve an efficient protection with real multipath routing, compared to unipath routing, is that an IP packet can be forwarded on several routes for a given destination even before any failure occurs. Hence, routes are not only designed for protection aspects but for load balancing too. We therefore cannot consider that, in alternate routes computation, a router forwards traffic only via the shorstest route. Conditions used to validate loop free alternate routes have to take into account the fact that routers may forward packets through different next hops for the same destination. In the first section we introduce a brief state of the art of most common techniques using multipath routing. Then, we present our distributed multipath routing technique DT(p). In the third section, we analyze, with several topologies, different IP routing techniques characteristics and advantages. We conclude with a discussion about future extensions to provide a global mechanism of restoration and protection.