Autotuning of parallel programs using the IBM Watsons Analytics data analysis system

А.Yu. Doroshenko, O.S. Novak, P.A. Ivanenko, A.M. Starushyk


In this paper an analytical model of the method of automatic adjustment of parallel programs (auto-tuning) is presented. The software implementation of this model is based on the formal transformations of code and using expert data as a foundation of the optimization process and further analysis of the results with the IBM Watsons Analytics system. The results of a practical experiment confirming the effectiveness of the approach used in optimizing parallel programs are presented. The principles of the IBM Watsons Analytics data analysis system are examined and the system itself is shown in action.

Problems in programming 2018; 1: 46-54


autotuning; parallel programs; IBM Watsons Analytics; optimization; statistical modeling


Federal plan for high-end computing: Report of the High-end computing revitalization task force (HECRTF) – Retrieved from

Automatic Parallelization with Intel Compilers – Retrieved from

Naono K., Teranishi K., Cavazos J., Suda R . Software automatic tuning from concepts to state-of-the-art results – New York: Springer, 2010. – 240 р.

Asanovic K. The Landscape of Parallel Computing Research: A View From Berkeley. Technical Report, University of California, Berkeley, 2006

Doroshenko A.Yu., Ivanenko P.A., Novak O.S., Hybrid model of autotuning that use statistical modeling. Problems in Program-ming. 2016, N 4. P. 27–32.

IBM Watson Analytics – Retrieved from

Analytics of data-mining and data-science methodology – Retrieved from

Watson Analytics Developer Center – Retrieved from

Fisher r-to-t test – Retrieved from



  • There are currently no refbacks.