FAW: Graphical Specification Techniques for Parallel Programming

点击翻译按钮获取中文摘要

When programming massively parallel computations, programmers must specify large numbers of processes and their interactions, efficiently map their logical structures onto disparate hardware structures, and debug in the absence of reproducibility and consistent global states. Some of these activities can be supported by extensions of sequential tools, but most require development of novel techniques specific to parallelism. This development is made even more challenging by the diversity of target architectures, the relative lack of experience with parallelism, and the range in degree and granularity of available parallelism. This research focuses on support for message-passing architectures. During the funding period, work will continue on parallel program specification and debugging, and begin on programming support specific to the Connection Machine. The abstractions provided by a programming environment determine a programmer's effectiveness in implementing and debugging algorithms, yet few abstractions support massive parallelism. A graph grammar- based editor can facilitate the concise, scalable specification of parallel programs as annotated graph families. Specification, however, is not enough. Program graphs must be available to other tools - such as debuggers, animators, and mappers - within the environment. This work will investigate the applicability of existing techniques, to design methods of partial visualization, and determine the feasibility and utility of lazy graph generation. It is intended to explore the adaptation of these graphical specification techniques for the Connection Machine.