I've thought about this as well for a while. Arc's approach seems to be flat files so far, which means any move to multiple servers is terra incognita on some level. I mean to build a nosql system at some point, but I want to do something besides yet another project. I want to understand why we must couple technology stack choices with where we want to be in CAP space. Why can't we have a single project that lets us tweak one knob for RAM vs persistent store, strong vs eventual consistency, and so on? Laying out the design space in one place may give us some chance at least accumulating lessons as we reinvent the wheel.