Maintaining Routing Structures under Deletions via Self-Pruning

TL;DR

This paper introduces a family of self-pruning expander graphs that maintain efficient routing structures under adversarial edge deletions, combining expander routing and pruning in an online, worst-case manner.

Contribution

It presents a novel family of expanders that are both easy to route and self-pruning, with a worst-case, online algorithm for maintaining these properties under edge deletions.

Findings

The self-pruning expander family supports efficient routing after deletions.

The pruning process is worst-case, limiting deletions per adversarial step.

Routing paths can be tightly controlled in length and congestion.

Abstract

Expanders are powerful algorithmic structures with two key properties: they are a) routable: for any multi-commodity flow unit demand, there exists a routing with low congestion over short paths, where a demand is unit if the amount of demand sent / received by any vertex is at most the number of edges adjacent to it. b) stable / prunable: for any (sequence of) edge failures, there exists a proportionally small subset of vertices that can be disabled, such that the graph induced on the remaining vertices is an expander. Two natural algorithmic problems correspond to these two existential guarantees: expander routing, i.e. computing a low-congestion routing for a unit multi-commodity demand on an expander, and expander pruning, i.e., maintaining the subset of disabled vertices under a sequence of edge failures. This paper considers the combination of the two problems: maintaining a routing for a unit multi-commodity demand under pruning steps. This is done through the introduction of a family of expander graphs that, like hypercubes, are easy to route in, and are self-pruning: for an online sequence of edge deletions, a simple self-contained algorithm can find a few vertices to prune with each edge deletion, such that the remaining graph always remains an easy-to-route-in expander in the family. Notably, and with considerable technical work, this self-pruning can be made worst-case, i.e., such that every single adversarial deletion only causes a small number of additional deletions. Our results also allow tight constant-factor control over the length of routing paths (with the usual trade-offs in congestion and pruning ratio) and therefore extend to constant-hop and length-constrained expanders in which routing over constant length paths is crucial.