Dense-Node Anti-Pattern

Dense-Node Anti-Pattern

Why a single Country node connected to 100M Person nodes wrecks every traversal.

Why it matters

Traversal cost is O(degree). A 'God' node becomes a hub through which every query has to walk — even queries that have nothing to do with countries.

Going deeper

Why dense nodes are so toxic to traversals: when Cypher expands a relationship, it iterates the relationship chain of the source node. A node with 100M outgoing edges has a 100M-long chain. Any traversal through that node — even one looking for a single specific neighbour — pays a large fraction of that cost unless there's a relationship-property index helping it skip ahead.

Two refactors, in order of cost:

1. Intermediate nodes (Region, Era, Bucket) that split the fan-out into smaller, more selective hops. Cheap, no Cypher rewrite at the leaves.
2. Relationship type/property split — use distinct relationship types (e.g. :RESIDES_IN_EU, :RESIDES_IN_US) so the traversal can ignore the non-matching chain entirely. Useful when intermediate nodes don't fit the domain.

A dense node is a single roundabout with 200 streets feeding into it.

Every commuter — even the ones just trying to pass through town — has to crawl through the same intersection. Traffic backs up not because the roads are bad but because one node owns the fan-out for the entire city.

The civic fix is the same as the graph fix: insert district roundabouts between the giant one and the streets. Each district handles a slice of the traffic; the original square only sees one inbound road per district. Cypher: replace (:Country)-[:HAS_CITIZEN]->(:Person)×100M with (:Country)-[:HAS_REGION]->(:Region)-[:HAS_CITIZEN]->(:Person).