Hi Chris, Neo4j should be suitable for your problem. I just built a similar application (using an email corpus for relationships not facebook).
The size of the system was the following: Nodes: 100,030,002 Edges: 3,041,030,000 Attributes: 140,120,000 We did standard friend-of-a-friend and other navigational queries. Most of our queries came back under 50 milliseconds (often under 10 milliseconds). In an EC2 instance we were able to ingest the entire 95GB graph in 100m39.661s. I don't think we did a really good job optimizing Neo4j or the VMs, plus it was all virtualized in EC2 (dedicated hardware would have been faster). But with those speeds and at that scale, we were happy without spending any time optimizing. :) -Todd On Mon, Apr 19, 2010 at 2:21 PM, snacktime <snackt...@gmail.com> wrote: > Would Neo4j scale to handle the friend relationships handled by Facebook? > Where I work we run a number of large social games mostly on Facebook, and > for a number of reasons we end up caching a lot of the friend data (mostly > because their api can be unreliable and slow). I'd like to have a more > intelligent caching system where our side is relationship aware, so when > relationships change we can show that to users, as opposed to doing stuff > like caching your friends list for 48 hours, regardless of whether it's > changed. > > Is Ne04j even in the ballpark for something like this? > > Chris > _______________________________________________ > Neo mailing list > User@lists.neo4j.org > https://lists.neo4j.org/mailman/listinfo/user > _______________________________________________ Neo mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user
