We present DASH, a C++ template library that offers distributed datastructures and parallel algorithms and implements a compiler-free PGAS(partitioned global address space) approach. DASH offers many productivity andperformance features such as global-view data structures, efficient support forthe owner-computes model, flexible multidimensional data distribution schemesand inter-operability with STL (standard template library) algorithms. DASHalso features a flexible representation of the parallel target machine andallows the exploitation of several hierarchically organized levels of localitythrough a concept of Teams. We evaluate DASH on a number of benchmarkapplications and we port a scientific proxy application using the MPI two-sidedmodel to DASH. We find that DASH offers excellent productivity and performanceand demonstrate scalability up to 9800 cores.
展开▼
