[Paper Review] Xheal: Localized Self-healing using Expanders

Gopal Pandurangan, Amitabh Trehan|arXiv (Cornell University)|Apr 5, 2011

Modular Robots and Swarm Intelligence23 references4 citations

TL;DR

Xheal is a fully distributed, localized self-healing algorithm for reconfigurable networks that maintains strong global network properties—such as expansion, spectral gap, and stretch—after repeated adversarial node insertions and deletions. It achieves this by using expander graphs (H-graphs) and localized repair operations that ensure bounded degree increase (O(κd)) and O(log n) stretch, while preserving connectivity and expansion through amortized O(κ log n) cost per repair.

ABSTRACT

We consider the problem of self-healing in reconfigurable networks (e.g. peer-to-peer and wireless mesh networks) that are under repeated attack by an omniscient adversary and propose a fully distributed algorithm, Xheal that maintains good expansion and spectral properties of the network, also keeping the network connected. Moreover, Xheal does this while allowing only low stretch and degree increase per node. Thus, the algorithm heals global properties while only doing local changes and using only local information. Our work improves over the self-healing algorithms 'Forgiving tree'[PODC 2008] and 'Forgiving graph'[PODC 2009] (using a similar model) in that we are able to give guarantees on degree and stretch, while at the same time preserving the expansion and spectral properties of the network. These repairs preserve the invariants in the following sense. At any point in the algorithm, the expansion of the graph will be either `better' than the expansion of the graph formed by considering only the adversarial insertions (not the adversarial deletions) or the expansion will be, at least, a constant. Also, the stretch i.e. the distance between any pair of nodes in the healed graph is no more than a $O(\log n)$ factor. Similarly, at any point, a node $v$ whose degree would have been $d$ in the graph with adversarial insertions only, will have degree at most $O(κd)$ in the actual graph, for a small parameter $κ$. We also provide bounds on the second smallest eigenvalue of the Laplacian which captures key properties such as mixing time, conductance, congestion in routing etc. Our distributed data structure has low amortized latency and bandwidth requirements.

Motivation & Objective

• To address the challenge of maintaining global network invariants—like expansion, connectivity, and low stretch—in dynamic, adversarially attacked networks.
• To design a self-healing mechanism that operates using only local information and localized actions, avoiding centralized control.
• To ensure that node degree increases are bounded by a small factor κ, even after repeated adversarial deletions.
• To maintain spectral properties (e.g., spectral gap) crucial for routing, mixing time, and congestion, despite topology changes.
• To improve upon prior self-healing algorithms like Forgiving Tree and Forgiving Graph by providing formal guarantees on stretch and degree, not just connectivity.

Proposed method

• The algorithm uses H-graphs—random κ-regular graphs that form expanders with high probability—to build and maintain primary and secondary expander clouds.
• After each adversarial deletion, the system performs localized repairs: inserting new edges or reconfiguring existing ones using local knowledge and leader-based coordination.
• When multiple expander clouds need to be merged due to insufficient free nodes, a leader is elected via BFS over O(log n) hops, and a new H-graph is constructed and broadcasted to restore the expander property.
• The repair process is amortized over prior deletions: costly merging operations are balanced by cheaper, earlier repairs, ensuring O(κ log n) amortized time and message complexity.
• The algorithm maintains probabilistic guarantees on expansion by reconstructing H-graphs when a cloud loses half its nodes, using union bounds to ensure high probability of expansion.
• The system uses a hierarchical structure with primary and secondary clouds to manage free nodes and enable efficient insertion and repair under dynamic conditions.

Experimental results

Research questions

• RQ1 Can a self-healing algorithm maintain edge expansion and spectral properties in a distributed, localized manner under repeated adversarial node deletions and insertions?
• RQ2 What is the minimal degree increase required to preserve global network properties like stretch and expansion while ensuring connectivity?
• RQ3 How can expensive topological repairs (e.g., merging clouds) be amortized over time to maintain low average cost?
• RQ4 Can localized repairs preserve O(log n) stretch and constant expansion even when the adversary deletes nodes strategically?
• RQ5 What is the trade-off between maintaining strong spectral properties and controlling node degree growth in dynamic, self-healing networks?

Key findings

• The network maintains edge expansion that is either as good as the expansion in the graph with only adversarial insertions or at least a constant, ensuring robust connectivity.
• The stretch (distance between any two nodes) is bounded by O(log n) times the stretch in the graph with only insertions, preserving efficient routing.
• Node degree increases by at most a factor of O(κ), where κ is a small parameter, ensuring scalability and load balancing.
• The amortized time and message complexity of repairs is O(κ log n), making the algorithm efficient for large-scale networks.
• The second smallest eigenvalue of the Laplacian (which governs mixing time and congestion) is preserved up to constant factors, ensuring good performance in routing and load balancing.
• The algorithm successfully maintains global invariants using only local information and actions, improving upon prior work that lacked guarantees on stretch and degree.