News Archives

RSS Feed

ParoC++: An Objectoriented model for adaptive HPC on the Grid

October 7, 2004

Date: Thursday October 7, 2004
Time: 11am-12:15pm
Location: Woodward 149

Dr. Pierre Kuonen < Pierre.Kuonen@eif.ch>
University of Applied Sciences of Fribourg (EIA-FR) Switzerland

Abstract: Adaptive utilization of resources in a highly heterogeneous computational environment such as the GRID is a difficult question. In this talk we will present an object-oriented approach to the solution using requirement-driven parallel objects. Each parallel object is a self-described, shareable and passive object that resides in a separate memory address space. The allocation of the parallel object is driven by the constraints on the resource on which the object will live. A new parallel programming paradigm is presented in the context of ParoC++ - a new parallel object oriented programming environment for high performance distributed computing. ParoC++ extends C++ for supporting requirement-driven parallel objects and a runtime system that provides services to run ParoC++ programs in distributed environments.