Promising approaches to developing scalable applications for hybrid supercomputers

This article studies the matter of developing scalable applications to be run on supercomputers built with up-to-date as well as just emerging techologies. Authors describe a number of problems concerning complicated memory hierarhy arising mainly due to using accelerators, i.e. architecture hybridization. A short review of existing approaches to overcome these problems is given. From the whole range of possible solutions there is selected the most promising area, which is described in more details.