Paper Details

Communicating Process Architectures (CPA)
 Title: Scaling PyCSP
 Conference: Communicating Process Architectures 2013
 Authors: Rune Møllegaard Friborga, John Markus Bjørndalenb, Brian Vinterc
(a) Bioinformatics Research Centre, University of Aarhus
(b) Department of Computer Science, University of Tromsø
(c) Niels Bohr Institute, University of Copenhagen
 Abstract: PyCSP is intended to help scientists develop correct, maintainable and portable code for emerging architectures. The library uses concepts from Communicating Sequential Processes, CSP, and is implemented in the Python programming language. This paper introduces a new channel implementation and new process types for PyCSP that are intended to simplify writing programs for clusters. The new processes and channels are investigated by running 3 benchmarks on two separate clusters, using up to 512 CPU cores. The results show that PyCSP scales well, with close to linear scaling for some of the benchmarks. 
Files:


BibTeX Entry


Full paper


Presentation