DocumentCode :
2186737
Title :
MDMP: Managed Data Message Passing
Author :
Jackson, Andrew ; Strand, Par
Author_Institution :
Plasma Phys. & fusion energy Earth & Space Sci., Chalmers Univ., Goteborg, Sweden
fYear :
2013
fDate :
23-26 Sept. 2013
Firstpage :
496
Lastpage :
502
Abstract :
MDMP is a parallel programming approach designed to provide users with an easy way to add parallelism to programs, optimise scientific simulation algorithms, and providing optimised communications to MPI-based programs without requiring them to be re-written from scratch. MDMP uses directives to allow users to specify what communications should take place in the code, and then implements those communications in an optimal manner using both the information provided by the user and data collected from instrumenting the code and gathering information on the data to be communicated at runtime. In this paper we outline the basic concepts and functionality of MDMP and discuss the performance that can be achieved using our prototype implementation of MDMP a range of benchmark cases.
Keywords :
message passing; parallel programming; program compilers; MDMP functionality; MPI-based programs; code compiler; managed data message passing; optimised communications; parallel programming; parallelism; scientific simulation algorithms; Benchmark testing; Delays; Libraries; Optimization; Parallel programming; Program processors; Runtime;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Symbolic and Numeric Algorithms for Scientific Computing (SYNASC), 2013 15th International Symposium on
Conference_Location :
Timisoara
Print_ISBN :
978-1-4799-3035-7
Type :
conf
DOI :
10.1109/SYNASC.2013.71
Filename :
6821188
Link To Document :
بازگشت