This paper addresses the problems associated with finding efficient means of representing, storing, and processing array data structures when these structures are accessed within loops in a dataflow system.

The author describes what he refers to as the “array problem” and goes on to identify four criteria that must be used to judge solutions to this problem. He then presents a survey and analysis of current methods used to represent and process such arrays. Initially, randomly accessed arrays are considered and three methods are examined: directed acyclic graphs (heaps); I-Structures; and a method referred to as the static-VAL approach. In the next section, which deals with sequentially accessed structures, the material is divided into three areas which cover static (or stored) streams, dynamic streams, and the University of Manchester approach (which suggests the use of a preserve/defer function), respectively.

The paper goes on to present two new solutions to this problem, the Macro-Actor and Token-Relabelling schemes, which the author claims overcome the many drawbacks associated with existing solutions.

Overall, the paper provides a balanced survey and unbiased criticism of existing solutions to a nontrivial problem. The new methods described in the paper do seem to offer solutions to certain areas of the problem; but, as the author says, the complexity of the problem of array handling in dataflow systems precludes any universal solution.

To conclude, this is a very useful paper which provides a valuable reference for anyone working in the field. The only criticism that may be leveled at the paper is the relatively small example chosen for the “application study.” However, the fact that a number of schemes have to be compared does provide a reasonable excuse for this selection.