Arango isn’t the clear winner in all of these. Unconfigured PG did pretty good. PG with JSON and no indexes did well too. It also held its own using less memory. Finally, PG has transactions where as Arango doesn’t really if you have an index constraint on a document attribute. This shows Arango is better than Neo4j and Mongo for graph resulted service and document store. It skipped PG’s recursive capability that probably allows for basic graph traversal to a known depth.