Topology Splitting
Network Partition
A network partition attack splits the network into isolated subgroups by cutting bridge edges between them. Unlike eclipse attacks that target a single node, partitioning affects entire regions of the network.
Partitions break consensus — each side may independently accept conflicting state, leading to chain forks and inconsistent views.
Connected Network
A healthy peer-to-peer network where all nodes can reach each other through multiple paths. The mesh topology provides redundancy — even if some links fail, messages can route through alternative paths.
Attacker Identifies Cut Points
The attacker maps the network topology and identifies bridge edges — links whose removal would disconnect parts of the network. In a DHT, this can be done by observing routing table entries through the Identify protocol and FIND_NODE queries.
Edge Cutting Begins
The attacker begins severing cross-partition connections by interfering with the transport layer — sending RST packets, exploiting NAT timeouts, or using malicious relay nodes to drop traffic. Some cross-partition links are broken.
Network Partitioned
All cross-partition edges are severed. The network is now split into two isolated subnetworks. Peers in Partition A cannot discover, route to, or communicate with peers in Partition B. Consensus breaks — each partition may fork independently.
Defense: Redundant Connectivity
libp2p defends against partitioning through: (1) Multiple transport protocols (TCP, QUIC, WebSocket) — harder to cut all paths; (2) Circuit Relay v2 — peers can route through relay nodes to bridge partitions; (3) AutoNAT — detects reachability changes; (4) Peer diversity — Connection Manager prioritizes peers from different networks/ASNs.