PAWS is a collection of software tools. Its fundamental goal is to try to evaluate the behavior of one or more application programs on a variety of parallel computer architectures, without performing an actual port.

PAWS consists of three distinct tools. The application characterization tool performs a dataflow analysis of the application (currently written in Ada). It maps the original code to an intermediate form (IF1), which was originally developed for the SISAL dataflow language. In this mapping process, the classical problems encountered in mapping an imperative language onto a single assignment model of computation are solved. The computer architecture characterization tool is basically a database containing performance figures on parallel computer architectures with respect to computational speed, interconnection network characteristics, control, and I/O characteristics. It contains both static data and analytical models that allow one to compute a performance figure starting from user input data. The final tool, which combines the results of both previous tools, is the Performance Assessment Tool; it allows the user to determine an application profile under ideal circumstances and on a given architecture (that is, with a specific process-to-processor mapping). The operation of PAWS is illustrated by two examples: a matrix multiplication and a numeric integration.

The basic purpose of this publication is to present research. It fulfills its purposes and is of the right length. Its best feature is its clarity, while its worst feature is the lack of some more technical detail (such as how Ada programs can be mapped onto the SIMD Connection Machine architecture). The physical form is appropriate.

The intended audience is people interested in parallel computing; readers will need some experience in porting an application to this type of machine. The references could have been extended somewhat, for example in the direction of scientific research in dataflow computer architectures and languages.