Architecture, Implementation and Performance Optimization in Organizing Parallel Computations for Simulation Environment

This paper discusses architectural concepts, implementation details and performance optimization techniques in the context of instrumental environment ISMA2015 supporting parallel computations for hybrid models. The paper considers the approach of organizing computations so that the user can work with the environment in the terms of the application field omitting the complex implementation details and to simply running models in a suitable mode: sequential, parallel on a multi-core machine or a cluster. The technology of the remote class loading is proposed. The framework for extending the library by new numerical methods is considered. The results of performance optimization are given. The technology of optimizing communication between cluster nodes is described. Simulation results are presented on the example of generated reaction-diffusion problems.