Message Passing and PGAS Exascale Concepts in EPiGRAM Applications

Dnr:

SNIC 2017/1-2

Type:

SNAC Medium

Principal Investigator:

Stefano Markidis

Affiliation:

Kungliga Tekniska högskolan

Start Date:

2017-01-03

End Date:

2018-02-01

Primary Classification:

10201: Datavetenskap (= Datalogi)

Webpage:

http://www.epigram-project.eu/

Allocation

Abstract

We investigate new features that are in available in MPI3.1 (non-blocking collectives, sparse collectives, new performance monitoring interface) and GPI (a PGAS library) in the two pilot EPiGRAM applications: iPIC3D and Nek5000. In particular, we study these features in large simulation runs to analyze scalability both in execution time and memory consumption. The EPiGRAM project focuses on the scalability of Message Passing and PGAS programming models, ad this allocation time will allows us to study scalability of these approaches towards exascale. The focus of this work is on a dramatic improvement of the communication kernel of the applications (iPIC3D and Nek5000) by using new concepts in programming models.