Yes, that's right. The control is centralized, but the storage and the computation are fully distributed. We spent a long time working on what we think will be

Yes, that's right. The control is centralized, but the storage and the computation are fully distributed. We spent a long time working on what we think will be a scalable architecture. The query language is deliberately simple to make everything fast (there's only B-tree lookups and set ops on object IDs) and easily parallelizable. And this is the 3rd implementation :-)