An Architecture for Using Multiple Communication Devices in a MPI Library

Authors:
Hernâni Pedroso;João Gabriel Silva
Affiliations:
-;-
Venue:
HPCN Europe 2000 Proceedings of the 8th International Conference on High-Performance Computing and Networking
Year:
2000

Citing 6
Cited 2

Monitors, messages, and clusters: the p4 parallel programming system

Parallel Computing - Special issue: message passing interfaces
A high-performance, portable implementation of the MPI message passing interface standard

Parallel Computing
Efficient layering for high speed communication: the MPI over Fast Messages (FM) experience

Cluster Computing
Myrinet: A Gigabit-per-Second Local Area Network

IEEE Micro
SCI: Scalable Coherent Interface, Architecture and Software for High-Performance Compute Clusters

SCI: Scalable Coherent Interface, Architecture and Software for High-Performance Compute Clusters
WMPI - Message Passing Interface for Win32 Clusters

Proceedings of the 5th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface

The Implementation of One-Sided Communications for WMPI II

Proceedings of the 8th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface
Device level communication libraries for high-performance computing in Java

Concurrency and Computation: Practice & Experience

Quantified Score

Hi-index	0.00

Visualization

Abstract

Every year new intercommunication technologies emerge and the parallel computing libraries have to evolve to use these new technologies. Most of the existing libraries have the communication medium dependent code embedded in the core and rely on it for the management of the environment. This makes it difficult to adapt the libraries to those emerging technologies. A new architecture, implemented in the WMPI library, cleanly separates the communication device dependent code from the library core, making that adaptation much easier. Third party institutions (technology vendors and other research institutions) can thus easily adapt WMPI for any technology. Additionally, the architecture enables the concurrent use of any number of communication media by the WMPI library. The freedom is also given to the developer of the communication device dependent code to use or not dedicated threads for sending and/or receiving. This paper describes the main characteristics and rationale of that new architecture.