Finogenov O. Methods of managing dynamic analysis of complex objects in multiprocessor computing systems

The thesis is devoted to researching and developing computing management methods in dynamic analysis of complex objects using multiprocessor computing systems (MCS). The paper reviewed the current status of CAD and highlights the main stages of MCS usage by circuit design tools. The usage of modern tools to improve the reliability of solution of the object dynamic analysis tasks is proposed and the existing computing management methods in dynamic analysis are examined. It is shown that the existing relationships to manage computations based on the method of numerical integration on the basis of differences of higher orders are non-optimal and it is suggested to use a set of method orders and step sizes to improve reliability. The features of integration step increasing and their influence on the reliability of solving dynamic analysis tasks are considered, and it is proposed to make use of the decreasing step coefficients vector. Computation management methods are developed, which depending on the method ensure the integration step maximization and the local error or number of Newton iterations minimization. An increase in efficiency and reliability of the proposed computation management methods is proved experimentally. Data exchange schemes for multiprocessor systems with common and shared memory are developed and theoretical estimations of communication complexity and execution time are obtained. The possibility of a priori estimation of computational complexity is shown and the recommendations on using management methods depending on the characteristics and architecture of MCS are given. Testing of the effectiveness and time estimation of the computation management methods implemented in the Allted circuit design package is fulfilled on the NTUU "KPI" supercomputer. Keywords: dynamic analysis, computation management methods, parallel algorithms, MCS, CAD, simulation.