The Creation Of Intelligent Support Methods For Solving Mathematical Physics Problems On Supercomputers

An approach to creating methods and means of intelligent support for solving compute-intensive problems (CI problems) of mathematical physics on modern peta- and future exaflops supercomputers, containing millions and, eventually, billions of simultaneously running computational cores and providing an enormous degree of parallelism, is proposed. The relevance of the intelligent support of the process of solving problems at all stages - from setting a problem, selecting a method of numerical solution to choosing a supercomputer architecture and software implementation is substantiated. The proposed system of intelligent support is based on the ontology of computational methods and algorithms, the ontology of parallel architectures and technologies and uses decision rules to find the best possible approach to parallel solving a problem specified by a user, at all stages of its solution, up to choosing the best planning strategy for the computational process. The paper describes the concept of creating intelligent support for solving CI problems, using ontologies and inference rules. An example demonstrating the use of the proposed approach for solving a problem of astrophysics is presented.