Models and methods for the profiling and execution time estimation of workflows in supercomputing systems

The solution of complex engineering and scientific problems using distributed computing and supercomputer systems can often be efficiently implemented by organizing workflows that integrate the individual program components (grid generators, solvers, visualizers, multi-criteria optimization systems, etc.) to solve the end problem. In order to optimize the load ratio of such platforms, one can use pattern recognition methods to detect such workflows and profiles evaluation methods to estimate their performance and execution time on the provided computing resources. A review of existing methods for the intellectual profiling and workflow execution time estimation is presented. A new mathematical model of representation of a computing job as a marked-up weighted directed acyclic graph is proposed. The problem of task graph clustering and its mapping onto the computing environment is considered. The work was supported by the Russian Ministry of Education (government contract N 14.514.11.4106) under the Federal Program “Research and development on priority directions of scientific-technological complex of Russia for 2007–2013”.