Parallel Database Engine (parade) Final Report Parallel Language Support (section 2). We Have Parallel Programming (section 4). Motivated by 2 Parallel Language Support

1 Summary This report summarises the achievements of the EPSRC Parade project (GRJ/53348), which ran from June 1994 to September 1997. We have made contributions in the following three areas. implemented a robust, portable runtime system and compiler support for Glasgow Parallel Haskell. The system supports a complete parallel development environment providing an integrated simulator and sophisticated prooling tools in addition to physical implementations on a variety of parallel machines. Despite sustained research interest in parallel functional programming, our implementation is one of the rst to be made publicly available, and to be widely used outside the research group that developed it. Large-scale data-intensive parallel programming (Section 3). In collaboration with groups at Durham University and the Centre for Transport Studies, London we have developed and measured several large parallel data-intensive programs. We have demonstrated wall-clock speedups for both textbook and, more signiicantly, real problems on both distributed-memory and shared-memory architectures. our studies of large-scale data-intensive programs, we have developed evaluation strategies, a new programming abstraction that cleanly separates code describing the algorithm from that controlling the parallel behaviour. Evaluation strategies can be used to describe a wide range of control and data parallel programming paradigms, are user-extensible, can be nested or composed arbitrarily, and provide a simple framework for reasoning about parallel program behaviour. Based on our experiences with using evaluation strategies , we are developing a methodology for constructing large parallel functional programs based on a combination of simulation, prooling and performance measurement on real machines. A major focus of this project is its emphasis on parallel programming rather than parallel implementations. In sharp contrast to the majority of projects in this area we have progressed beyond simply obtaining a working parallel implementation to studying the problems associated with producing real parallel programs written in a functional style, and manipulating large quantities of data. We have published widely, producing more than 20 Journal , Conference, and Workshop papers during the course of the project, edited proceedings and contributed to several books. Several other papers are in preparation. We interact with many groups both nationally and internationally, and have recently been instrumental in establishing a special interest group within Scotland. Encouraged by free, public availability, our software and techniques are being used and further developed by groups worldwide. In the following text, project publications are cited numerically , thus 12], and are listed at the end of each section. …