Recursive structure in computer systems

Abstract

Structure plays a important part in the design of large systems. Unstructured programs are difficult to design or test and good structure has been recognized as essential to all but the smallest programs. Similarly, concurrently executing computers must co-operate in a structured way if an uncontrolled growth in complexity is to be avoided. The thesis presented here is that recursive structure can be used to organize and simplify large programs and highly parallel computers. In programming, naming concerns the way names are used to identify objects. Various naming schemes are examined including 'block structured' and 'pathname' naming. A new scheme is presented as a synthesis of these two combining most of their advantages. Recursively structured naming is shown to be an advantage when programs are to be de-composed or combined to an arbitrary degree. Also, a contribution to the UNIX United/Newcastle Connection distributed operating system design is described. This shows how recursive naming was used in a practical system. Computation concerns the progress of execution in a computer. A distinction is made between control driven computation where the programmer has explicit control over sequencing and data driven or demand driven computation where sequencing is implicit. It is shown that recursively structured computation has attractive locality properties. The definition of a recursive structure may itself be cyclic (self-referencing). A new resource management ('garbage collection') algorithm is presented which can manage cyclic structures without costs proportional to the system size. The scheme is an extension of 'reference counting'. Finally the need for structure in program and computer design and the advantages of recursive structure are discussed.