Algorithms for Planning on Computational Model with Redundancy and Uncertainty

Nowadays, the development of scalable scientific and applied applications is one of the key challenges for modern distributed environments that include heterogeneous computing resources. Clusters of personal computers, high-performance systems of public access supercomputer centers, or cloud platforms often represent these resources. Usually, cluster nodes of different types differ significantly in their computational characteristics (the number of nodes and cores, RAM and disk memory, interconnect, system software, administrative policies, job scheduling principles, and many other parameters). Therefore, planning computations in distributed applied software packages is an important problem. To this end, we propose new models and algorithms for computation planning in software packages. Their distinctive feature is the analysis of the execution time of package modules in the environment nodes. We obtain such time evaluations in the process of their continuous integration, delivery, and deployment. The proposed models and algorithms allow us to create redundant problem-solving schemes. In practice, the problem-solving scheme redundancy enables us to adapt it to the changing characteristics of environment resources and improve the computation reliability. Construction of problem-solving schemes is shown in model examples.