If I'm reading the high-level description correctly, the plan here is for this thing to be one giant centralized instance. Can anyone verify that I'm reading that right? Not that I have any pro or con feelings towards this architecture, just not sure if I'm fully grokking it.
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 :-)
