PVM: a framework for parallel distributed computing
Concurrency: Practice and Experience
Component Software: Beyond Object-Oriented Programming
Component Software: Beyond Object-Oriented Programming
Accurately Measuring MPI Broadcasts in a Computational Grid
HPDC '99 Proceedings of the 8th IEEE International Symposium on High Performance Distributed Computing
MPICH-G2: a Grid-enabled implementation of the Message Passing Interface
Journal of Parallel and Distributed Computing - Special issue on computational grids
A Network Topology Description Model for Grid Application Deployment
GRID '04 Proceedings of the 5th IEEE/ACM International Workshop on Grid Computing
On Abstractions of Software Component Models for Scientific Applications
Euro-Par 2008 Workshops - Parallel Processing
Implementing endogenous and exogenous connectors with the common component architecture
Proceedings of the 2009 Workshop on Component-Based High Performance Computing
| Hi-index | 0.00 |
Existing high performance component models mainly focus on the efficiency of the composition of two components, tackling also the special case of parallel components that enable N x M communications. The implementation of parallel components is usually assumed to be done thanks to some external communication paradigms like MPI. However, as of today, collective communication operations like broadcast, reduction, gather, etc. are not supported by component models. Programmers should develop such operations on top of point-to-point communication operations provided by component models. This paper studies how collective operations between components can be provided from an user and developer point of view. The result is an abstract component model that allows the implementation of collective communications. Software components are then able to use collective communications between several instances. To be effective on hierarchical resources such as grids, the model is hierarchical and relies on the concept of replicating component implementation. Last, the paper deals with the projection of such an abstract model onto existing models. It is validated through some very preliminary experiments.