A Suggested Architecture for the Massive Memory Machine
This report describes a computer architecture with two classes of primary storage; a tree-structured memory which is massive, but slow, and a standard, paged memory which is fast, but small. We then argue that only a small percentage of a program's memory references, those made to large (heap) data
structures, are responsible for the thrashing a massive memory machine is designed to avoid. Hence, mapping the heap into the tree-structured memory and placing program segments which are well-behaved in the standard memory retains the benefits of the massive memory, while minimizing the impact of its
slower access time. Finally, we propose several enhancements to the basic architecture that permit limited parallel processing.