In early 1969, BBN began an effort aimed at developing a new time-shared operating system. It was felt at the time that none of the commercially available systems could meet the needs of the research planned and in progress at BBN. The foremost requirement of the desired operating system was that it support a directly addressed process memory in which large list-processing computations could be performed. The cost of core storage prohibited the acquisition of sufficient memory for even one such process, and the problems of swapping such very large processes in a time-sharing environment made that solution technically infeasible as well.
The problem of the use of two levels of storage for programs is explored in the context of a LISP system which uses core memory as a buffer for a large virtual memory stored on a drum. Details of timing are given for one particular problem.
In an ideal list-processing system there would be enough core memory to contain all the data and programs. Described in this paper are a number of techniques that have been used to build a LISP system utilizing a drum for its principal storage medium, with a surprisingly low time penalty for use of this slow storage device. The techniques include careful segmentation of system programs, allocation of virtual memory to allow address arithmetic for type determination, and a special algorithm for building reasonably linearized lists. A scheme for binding variables is described which is good in this environment and allows for complete compatibility between compiled and interpreted programs with no special declarations.
: The report describes in detail the BBN-LISP system. This LISP system has a number of unique features; most notably, it has a small core memory, and utilizes a drum for storage of list structure. The paging techniques described allow utilization of this large, but slow, drum memory with a surprisingly small time penalty. These techniques are applicable to the design of efficient list processing systems embedded in time-sharing systems using paging for memory allocation.
