It is clearly crucial for the success of object—oriented databases to find efficient implementations that improve on the performance of relational systems, rather than being powerful in terms of modeling and features, but just too slow to be used. This paper describes the mapping of COCOON to DASDBS, a nested relational database kernel system, as an example OODBMS mapping to a complex storage system. We describe 1) choices for physical designs that make use of the complex storage model and 2) the generation of efficient, set—oriented execution plans for object—oriented database queries, using rule—based query optimization techniques. We use hierarchical clustering and embedded (sets of) object references, and show how to explore them for efficient path traversals expressed in queries involving complex objects. Prototypes of both, a physical design tool and a query optimizer have been implemented. Preliminary results show feasibility, and execution time improvements of an order of magnitude.