Two extra peers triple XRPL resilience in 2022 map study

An arXiv paper published Aug. 26 analyzed a set of 2022 XRP Ledger network snapshots and reported that small, random peer additions can raise modeled resistance to targeted node removals. The paper models how adding a fixed number of random peer connections per participating server changes which node removals are needed to break quorum connectivity.

The researchers reused 1,290 hourly snapshots collected over two months in 2022 and selected one representative graph for detailed experiments. That snapshot had 952 nodes, 15,070 edges and an average node degree of 31.7. The team ran Monte Carlo simulations that combined random topology change with targeted removals of nodes ranked by degree or by betweenness centrality.

The core experiment, called K-out augmentation, has each participating node create K new undirected links to peers chosen uniformly at random. At 60% participation with K=2, simulations where attackers removed the highest-degree nodes raised the modeled quorum-critical attack size from about 11% to 38%. When removals targeted nodes by betweenness centrality, the threshold rose from about 12% to 33% under the same participation and K values.

The metric used in the experiments measures the share of nodes that must be removed before fewer than 80% of the simulated validators remain together in a single connected component. That measure is a proxy for the network’s ability to keep a large fraction of validators connected for consensus.

The paper also compared random K-out augmentation to an edge-rewiring approach. At K=3 and at 80% or 100% participation, random augmentation produced robustness similar to or greater than roughly 20–25 iterations of rewiring, while preserving about 0.85 Jaccard similarity with the original edge set. Rewiring produced a Jaccard similarity of less than 0.5 in the authors’ tests. The authors released their simulation code and the snapshot files used for the experiments.

Validator placement was modeled separately because the dataset did not identify validators. Each main simulation selected 34 validators uniformly at random and excluded them from direct targeting. Sensitivity tests that concentrated validators on high-degree or low-degree nodes changed absolute thresholds but did not remove the relative benefit of random peer augmentation.

The paper notes that the experiments use a historical topology rather than a live mainnet snapshot. Public node counts available at a later date differed from the 2022 crawl used in the paper: a third-party node explorer showed 786 discoverable nodes and listed 35 validators in a visible UNL on Aug. 30. The authors point out blind spots in public crawlers: a peer crawler can omit IP address or port when a connected peer is a validator or a private peer, and official validator guidance encourages private or protected peer paths. Those factors limit recursive endpoint discovery and make a fresh topology measurement with defined coverage necessary for updated modeling.

The paper lists practical constraints that the graph model does not capture. Current xrpld software defaults to a soft maximum of 21 peers and maintains at least 10 outgoing connections. Raising the soft maximum to a value below 68 does not in itself increase outbound connections because of the incoming-to-outgoing allocation logic. Fixed peers, peer reservations and manual connections can exceed the soft maximum, but durable bilateral links require coordination between server operators. Increasing peer counts also raises bandwidth use. The simulations do not model operational costs or risks such as peer-slot contention, privacy trade-offs, exposure to malicious peers or denial-of-service effects.

The paper’s reported percentages are outputs from the modeled 2022 graph and the authors note that applying the same tests to the live network requires a current topology map, a defensible model for validator placement and an operational trial of how random links would be selected, accepted and maintained. Until such measurements and trials are run, the numerical thresholds remain tied to the 2022 snapshot used in the simulations.