>>But in answering this, I wonder if there are actually two use cases here
Yes, I see the use cases as the design decision points you are forced to make at varying points in the scale of increasing data volumes: 1) 0-10s of gigabytes: Slam in the RAM on a single server and all is plain sailing 2) Hundreds of Gigabytes Too big to hold all in RAM on a single server but not too big to worry about the cost of replicating the data on disk. Use the suggested "intelligent cache router" to favour replica servers with a likelihood of a pre-warmed cache for the given keys. The cost of a cache miss is not too catastrophic ( a local disk read vs RAM access) 3) Terabytes and above Too big for RAM, too big to store or replicate in its entirety on each server. The cost of not finding what you are after in RAM is then potentially very large - not just a local disk read but due to sharding potentially a network hop and related issues of the traversal state must now be exchanged between server processes. Cheers Mark _______________________________________________ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user