Distributed systems commonly contain heterogencity at all lcvels of systems structure, differentiated by function (special servers), operating systems and architecture within a single system. On the other hand, large mainframes tend to be more homogeneous in their structures, even when they are multiprocessors. This paper explores a way of using the models of heterogeneous distributed computing within a mainframe. The argument is that appropriate restructuring of the mainframe can achieve a convergence of network oriented and mainframe oriented models of computing and show that a mainframe is just a special case of a distributed system that happens to exist in a box rather than be distributed across wider distances. While there are different views about how many of the "aspects" of distributed processing such a system would have, it would potentially have the advantage of putting large frames on the same power and price/performance curves as collections of distributed nodes. And it would add some attractive features to mainframes.A key concept is that computer structures scale up and down an communications becomes less of a barrier. When looking at a map of a coastline, it is necessary to have reference points to know whether the shape is a few miles, a few hundred miles or a few thousand miles. The analogy for computing is that in the future it will be true that in regarding a population of processors one will need a reference point to know whether the population exists within a frame, within a room, within a building, or spans greater geographical distances.A new structure for mainframes and their relationships to endpoint devices is presented. The author feels that this structure reflects a realistic interpretation of a number of technology trends and information management concerns.It is tremendously important for the reader to remember that a paradigm of computing dominating thinking at the beginning of a decade need not be the paradigm that actually emerges, Historically we move quickly from one pardigm to another. It took just 5 years to move from concepts of highly centralized MULTICS driven computer utilities to complete decentralization of computing, and on to various images of distributed processing. Current paradigms that stress agglomerations of small machines may be just as ephemeral and our notions of where to place enduser interface functions may well not yet be mature.