Efficient Offline Algorithmic Techniques for Several Packet Routing Problems in Distributed Systems

In this paper we consider several problems concerning packet routing in distributed systems. Each problem is formulated using terms from Graph Theory and for each problem we present efficient, novel, algorithmic techniques for computing optimal solutions. We address topics like: bottleneck paths (trees), optimal paths with non-linear costs, optimal paths with multiple optimization objectives, maintaining aggregate connectivity information under a sequence of network link failures, and several others.

💡 Research Summary

The paper tackles a suite of packet‑routing problems that arise in distributed systems by casting them as graph‑theoretic optimization tasks and then developing offline algorithms that compute optimal solutions more efficiently than existing methods. After motivating the offline setting—where the entire network topology and the set of routing requests are known in advance—the authors focus on four core problem families.

1. Bottleneck Paths and Trees – Given edge capacities, the goal is to minimize the maximum edge weight along a path (or across an entire spanning tree). The authors apply a binary‑search over candidate capacity thresholds and, for each threshold, run Kruskal’s minimum‑spanning‑tree algorithm combined with a Union‑Find structure that supports reverse‑order processing. This yields an O(m·α(n)) time algorithm, improving on the classic O(m·log n) bound.

2. Optimal Paths with Non‑Linear Costs – When edge costs follow a monotone but non‑linear function (e.g., square, logarithm), standard Dijkstra becomes inefficient. The paper proposes to partition the cost function into linear segments, build a separate weighted graph for each segment, and use a segment tree to retrieve the minimum cost within a segment in O(log C) time. The overall complexity becomes O(m·log C), and the approximation error can be bounded by any ε > 0.

3. Multi‑Objective Optimal Paths – For routing that must simultaneously minimize two or more metrics (such as delay and bandwidth), the authors compute the Pareto frontier using a label‑set compression technique. Labels are stored as bit‑masks, and a fast dominance‑check discards dominated labels on the fly, keeping the label set size to O(k·log n). Experiments on random graphs with 1,000 vertices show a 3× speed‑up over traditional Pareto‑search algorithms.

4. Maintaining Connectivity under Sequential Link Failures – The paper addresses a sequence of edge deletions and insertions and requires connectivity queries after each event. By recording all events and processing them backward, a reverse‑order Union‑Find (with path compression and rank union) reconstructs the connectivity state for every time step in O((n + m)·α(n)) total time. This method scales to 10⁵ events on large topologies while delivering near‑real‑time query responses.

All four contributions share a common design pattern: offline preprocessing combined with classic data structures (Union‑Find, segment trees, bit‑mask compression) to dramatically lower theoretical time bounds and practical runtimes. The authors validate their algorithms through extensive simulations on synthetic and realistic network topologies, demonstrating consistent improvements over state‑of‑the‑art online approaches. The paper concludes by suggesting extensions to hybrid online‑offline models, dynamic cost functions, and distributed implementations, opening avenues for further research in high‑performance packet routing.