On ad hoc routing with guaranteed delivery

We point out a simple poly-time log-space routing algorithm in ad hoc networks with guaranteed delivery using universal exploration sequences.

💡 Research Summary

The paper addresses the fundamental problem of guaranteed message delivery in static ad‑hoc wireless networks where each node is severely constrained to O(log n) memory and can only attach O(log n) bits of overhead to messages. The authors show that, despite these harsh restrictions, it is possible to design a routing (and broadcasting) algorithm that always terminates in polynomial time and either delivers the message to the intended destination or correctly reports failure.

The cornerstone of the solution is the use of Universal Exploration Sequences (UES). A UES is a deterministic sequence of “directions” that, when followed on any connected 3‑regular undirected graph, is guaranteed to visit every vertex within a length that is polynomial in the number of vertices. Reingold’s breakthrough result (ST‑Connectivity in Log‑Space) provides a log‑space construction of such sequences, meaning that any element of the sequence can be computed on‑the‑fly using only O(log n) workspace.

To apply UES to arbitrary networks, the authors first perform a standard degree‑reduction transformation that converts the original graph into a 3‑regular graph G′. Each original vertex v of degree d is replaced by a small “gadget” of O(d) 3‑regular vertices, inflating the graph size only by a constant factor. The transformation is purely local, requiring no global knowledge.

Routing proceeds as follows. A message carries a header consisting of five fields: source identifier s, destination identifier t, a direction flag (forward or backward), a status bit (success/failure), and the current index i into the UES. Starting at s, the node reads the i‑th element t_i of the UES (computed in log‑space) and, using the locally stored port labeling, forwards the message to the neighbor dictated by t_i, then increments i. If the message reaches t, the direction flag is flipped to “back”, the status is set to success, and the message backtracks along the same UES (decrementing i) until it arrives at s, where the success is reported. If the index i exceeds the length L_n of the UES without encountering t, the algorithm concludes that t is not reachable from s; it flips the direction to “back”, sets status to failure, and backtracks to s to report the negative outcome. All computations at each hop involve only the current node’s incident edge labels and the O(log n) header, fitting within the memory budget.

A notable contribution is the method for determining the size of the connected component C_s of the source without any prior knowledge of n. The algorithm runs a sequence of UESes of lengths 2^k (for k = 1,2,…), each time checking whether every neighbor of the vertices visited so far is already in the visited set. When this condition holds, the visited set equals the entire component, and the algorithm can count its vertices by a second pass over the same UES. This counting procedure runs in time polynomial in |C_s| and uses only logarithmic space.

The main theorem proved is: there exists an ad‑hoc routing algorithm on a static network such that each node uses O(log n) space, the message overhead is O(log n), and the total running time is polynomial in the size of the source’s connected component. The same algorithm works for broadcasting. Moreover, any probabilistic routing algorithm with expected time T(n) and failure probability smaller than 1/poly(n) can be combined with the deterministic UES‑based algorithm to obtain a new algorithm with expected time O(T(n)) that is guaranteed to succeed whenever a path exists.

The paper acknowledges that the currently known constructions of UES have relatively high polynomial degree, which may limit practical efficiency. Nonetheless, the theoretical result is significant: it demonstrates that guaranteed delivery does not require linear memory or global topology knowledge, even in worst‑case 3‑dimensional graphs. Future work is suggested in designing shorter UES, optimizing the degree‑reduction gadget, and extending the approach to dynamic or directed networks.