DIVE is a distributed VE model that uses a shared dataspace approach. Considerable emphasis is laid here on implementation efficiency, using replication and multicast protocols. DIVE's dataspace has a tree structure, like VRML's or Amulet's. Nodes of the tree may have behaviour associated with them, which can be written in TCL. Data values (called `properties') may be written into the nodes, and read by other nodes. The system supports basic traversal through the tree using parent and child lookup functions. Change notification is done using events.
Something I found rather disappointing about the model is its untransparent handling of updates between the replicas of the database. Though updates may be delayed, no clear specification is given of the precise extent of inconsistencies between the replicas in different situations, and apparently no programming tools are provided to deal with this problem.