作者: Eric de Sturler , Milind Bhandarkar , L. V. Kalé
DOI:
关键词:
摘要: Parallel Computational Science and Engineering (CSE) applications often exhibit irregular structure dynamic load patterns. Many such have been developed using procedural languages (e.g. Fortran) in message passing parallel programming paradigm MPI) for distributed memory machines. Incorporating balancing techniques at the application-level involves significant changes to design of applications. On other hand, traditional run-time systems MPI do not support balancing. Object-based languages, as Charm++ efficient object migration applications, well deal with external factors that cause imbalance. However, converting legacy object-based paradigms is cumbersome. This paper describes an implementation MPI, called Adaptive (AMPI) supports multithreading Our approach based on user-level migrating threads capabilities provided by framework. Conversion from codes this platform straightforward even large codes. We converted component ROCFLO ROCSOLID a Rocket Simulation application AMPI. experience shows minimal overhead effort, one can incorporate Fortran-MPI