Control of resource-intensive computations under uncertainty. I. Multiparametric model

A complex of models, algorithms and programs designed to control a high-performance heterogeneous specialized computing system that executes heterogeneous tasks is presented. A multi-parametric model designed to control processing of resource-intensive heterogeneous computational tasks that can be parallelized with respect to data is given. The model relies on operation research technique and uses the principle of guaranteed result for control under objective uncertainty. During scheduling, maximin estimates of unknown variables (time in the system, processing duration, etc.) are calculated at each check point and used to form scheduling control commands. This approach allowed implementing a fundamentally new service discipline in practice that implies less resource-intensive tasks to be solved first.