Relational databases can run graph queries faster than dedicated graph engines.

Many people assume you need a specialized graph database to study connected data, like social networks. These tools store data as points called nodes and lines called edges. The authors argue that this separate setup is unnecessary and creates extra work. Regular relational databases, which organize data in standard tables with rows and columns, already track these links. To prove this, the researchers created ClickGraph and DeltaGraph. These systems take graph queries written in Cypher and turn them directly into standard SQL. The queries run right inside existing tools like ClickHouse or Databricks without importing data or setting up extra computers. The researchers found that regular columnar systems match or beat native graph engines on analytical questions. They also keep working at massive scales where in-memory graph systems completely run out of room and fail.

Why it matters

Enterprises often invest in specialized graph databases to analyze connected data, but this work shows they can use their existing relational systems for faster, scalable results without extra tools or data movement.

Questions answered