High level abstrations for parallel computing

点击翻译按钮获取中文摘要

Nowadays, even though there is an abundance of parallel computers around us in the form of multi-CPU PCs, shared-memory multiprocessors, and message-passing processor-networks, a majority of the programs are still being written for single processor computers. The current state of the art lacks parallel programming models and systems that are flexible, easy-to-use, and applicable to most of the common computer architectures. Researchers have been experimenting with high level parallel computing models that rely on new or extended languages, GUI based environments, pragmas, parallel patterns and skeletons for addressing these issues. This research proposes  a generic pattern based parallel programming model, called GPE, that: (a) supports parallel programming on shared-memory as well as message-passing computers; (b) is based on simple expressions, called Generic Pattern Expressions (GPEs), for instantiating common parallel structures and behaviors; (c) provides the required trade-off between parallelization effort and performance; (d) can be implemented in a common programming language (e.g., C++) without requiring any new language or language extensions; (e) supports user-written shared-memory or message-passing based new patterns. Our recent research work on parallel programming systems PAS and EPAS demonstrates that it is possible to design and implement an extensible pattern-library based parallel programming model and system while our work on Active Expressions shows the feasibility of instantiating common parallel behaviors without requiring any language extensions, preprocessing, or GUIs. Even though satisfying above-stated objectives (a) to (e) in a single research project may be a challenging proposition, it is worth pursuing and we believe, given certain recent advances in this research area, these may be feasible.