Skeletons and abstract models: how to get portability and performance in parallel programming

Lines connect the coordinator with its partners.

Project objective

Research objectives and content In the last years skeletons theory offered a simple and powerful approach to parallel programming. Unfortunately this methodology suffers of a serious lack of portability. The main goal of my proposal is to connect such a programming style and the of the abstract models for parallel computing. Models as BSP or LogP may be exploited implementing a skeletons language for the following reasons: - they provide a cost model to design a good compiler - they offer a portable way to implement the language supportPratically I will design a prototype of a BSP-based skeletons language trying of to provide: - new implementation scheme (template) - new optimization technique At the end of such a analisys a more suitable abstract model may be designed. Training content (objective, benefit and expected impact) The connection between skeletons programming and abstract models for parallel computing is the subject of my PhD thesis. The idea is partially due to my previous visit in Edinburgh as TRACS visitor and to my cooperation with Dr. Murray Cole. Developing my project in Edinburgh also means to be able of to exploit the large-scale computing facilities of the Edinburgh Parallel Computing Center. Links with industry / industrial relevance (22)P3L is a skeletons language designed by my group. At the moment P3L is the kernel idea of the SKIE environment adopted by QSW for a new parallel machine: PQE2000. QSW is an european joint venture between Alenia (Italy) and Meiko (UK).

Original text from CORDIS.

Participants

• UNIVERSITY OF EDINBURGH · EDINBURGHCoordinatorUnited Kingdom

Links

• View on CORDIS