New features for advanced dynamic parallel communication routines in OpenPALM: Algorithms and documentation

  • Martin Wlotzka (Author)
    Engineering Mathematics and Computing Lab (EMCL), Interdisciplinary Center for Scientific Computing (IWR), Heidelberg University
  • Thierry Morel (Author)
    Centre Europeen de Recherche et de Formation Avancee en Calcul Scientifique (CERFACS), Toulouse
  • Andrea Piacentini (Author)
    Centre Europeen de Recherche et de Formation Avancee en Calcul Scientifique (CERFACS), Toulouse
  • Vincent Heuveline (Author)
    Engineering Mathematics and Computing Lab (EMCL), Interdisciplinary Center for Scientific Computing (IWR), Heidelberg University

Abstract

We present an algorithmic sketch and documentation of the new dynamic distributors feature in the parallel communication routines of the OpenPALM software coupler tool. OpenPALM controls the execution of coupled applications and provides communication routines for data exchange between the units. A typical use case of OpenPALM is the coupling of numerical models for data assimilation or multiphysics simulations. The communication routines allow to transfer data between units exhibiting internal parallelization and data distribution. OpenPALM is able to match the distributed object parts between parallel units by means of its internal routing table. However, in the legacy OpenPALM version 4.1.4, the parallel configuration of the units and their data distribution needs to be determined through offline tests before the coupling application can run. Moreover, the data distribution cannot change anymore during runtime of the application. This may result in substantial overhead for the setup of applications, and it prevents from common practices in numerical models which may change data sizes and distributions during runtime like coarsening, refinement or load balancing. Our new developments allow to determine and change the data distribution among coupled units at any time during the execution of the coupling application, while at the same time keeping OpenPALM’s internal routing table consistent. This greatly eases the setup of applications and supports coarsening, refinement and load balancing during runtime.

Statistics

loading
Published
2017-06-13
Language
en
Academic discipline and sub-disciplines
OpenPalm, Model Coupling, HPC
Keywords
OpenPalm, Model Coupling, HPC