Resumen
Advances in the computational capabilities of high-performance architectures make it possible for computational scientists and engineers to address increasingly challenging problems. At the same time, it is becoming considerably more difficult to build software that achieves high performance on these systems. Programs addressing large-scale scientific problems often use sophisticated numerical techniques employing sparse or hierarchical data structures. Solving these important problems may require coupling several independently developed application codes, some using very large data sets consisting of empirical or simulated data. The five systems, Kernel Lattice Parallelism, Titanium, the Active Data Repository, superLU and Prometheus, are leading examples of emerging high-performance tools, languages, libraries and environments. They are described separately although they are not unrelated. All target high-end scientific computing application domains, taking advantage of characteristics of these domains to obtain better solutions than generic software tools.