Ah. We have progressed quite a bit beyond that paper.
We have a new all distributed abstraction (described on the website) in v2 based on a "gather/apply/scatter" model of update functions. We have an upcoming paper which will describe this system in greater detail.
The new implementation does not require a shared data table, using "syncs" to local variables to accomplish the same effect. We might introduce a DHT data-structure later on as a "support object" for the problems which require it.
Yucheng