Parallelization of a Scientific Application Using Didal Distributed Data Library

This chapter delves into the parallelization of the RHD3D application, which is used for numerical modeling of colliding flows in relativistic hydrodynamics. The focus is on using the Didal Distributed Data Library and fragmented programming technology to create an efficient, portable parallel program. The paper provides a detailed description of the fragmented programming technology, the architecture and design decisions of Didal, and the steps taken to parallelize the RHD3D application. Performance and efficiency testing results are presented, with a comparison to an existing Coarray Fortran parallel implementation. The paper also discusses the potential of Didal as a target platform for automatic program construction using the active knowledge concept. Readers will gain insights into the benefits of fragmented programming, the capabilities of the Didal library, and the performance improvements achieved through its use. The chapter concludes with a discussion on future research directions, including the introduction of high-level constructs to Didal to further simplify and speed up the development of parallel programs.