Local Fast Rerouting with Low Congestion: A Randomized Approach

Abstract

Most modern communication networks include fast rerouting mechanisms, implemented entirely in the data plane, to quickly recover connectivity after link failures. By relying on local failure information only, these data plane mechanisms provide very fast reaction times, but at the same time introduce an algorithmic challenge in case of multiple link failures: failover routes need to be robust to additional but locally unknown failures downstream. This paper presents local fast rerouting algorithms which not only provide a high degree of resilience against multiple link failures, but also ensure a low congestion on the resulting failover paths. We consider a randomized approach and focus on networks which are highly connected before the failures occur. Our main contribution are three simple algorithms which come with provable guarantees and provide interesting resilienceload tradeoffs, significantly outperforming any deterministic fast rerouting algorithm with high probability.

Improving the Resilience of Fast Failover Routing

“…For comparing schemes in terms of their hop count, we only take into account the runs where the EDPs succeeded. 2…”

“…Some of these fast failover algorithms leverage packet header modification, e.g., to carry failure information [21,28], by employing MPLS [31] or to leverage graph exploration approaches [4], but they can be disruptive to other network functions and moreover require specialized equipment or innode computation abilities, similar arguments can be made for mechanisms that require dynamic state on the forwarding devices, such as, e.g., resilience via a rotor router [11]. Recent work also investigated the use of randomization [2,7], but they require some means of random number generation on the routers and can lead to packet reordering.…”

Randomized Local Fast Rerouting for Datacenter Networks with Almost Optimal Congestion

“…It is resilient to O (k/ log k) link failures, while keeping the total load below O k L−1 log k • log log k w.h.p. For a certain class of routing protocols, that only forward over shortest paths (according to the local view of the nodes, cf Definition 22) and exhibit a property we call fairly balanced, this is again optimal up to (log log n) 2 factors. This class of protocols is natural and reminiscent of the widely-deployed shortest-path routing protocol ECMP (equal-cost multipath) [15,23].…”

“…The main algorithm combines the advantages of deterministic protocols and forwarding loop-freeness, with the resilience of randomized approaches. Our approach builds upon the Interval protocol in [2], which is designed for the clique. However, the adaptation of this approach to the Clos topology comes with multiple challenges that need to be solved.…”

“…In particular, we consider a scenario where n − 1 sources inject one indefinite flow each to a single destination. This scenario has already been studied intensively in the literature [2,3,10,22]: it models practically important operations such as in-cast [14,25], and is also theoretically interesting as it describes a particularly challenging situation because it is focused on a single destination which can lead to bottlenecks.…”