Tools, methods and algorithms for the efficient parallelization of computational loading in heterogeneous environments

The article is devoted to the current state analysis of research in the field of algorithmic, mathematical and software support for distribution tasks on compute nodes in a heterogeneous environment. Proposed new classification of load balancing strategies: on the dynamics principle, on the management basis, on the universality basis, with forecasting / without predicting of system status and others. Investigated some load balancing methods, systems and complexes, including the following: method of the problem submission into a directed acyclic graph, scheduler model based on the metadata, “DIET”, “ProActive”, “Moab”, “Maui”, support system for the moldable jobs, complex stream processing in the terms of queuing theory and service-oriented approach. It allow to minimize devices downtime-computing, reduce volume and transmission time for data from one device to another, improve overall scalability, minimize data access time and so on. Identified the advantages and disadvantages, presented use proposals. (In Russian).