Seminar Leader: Sadovnichy V.A., academician, rector of Lomonosov Moscow State University

The seminar was organized on the basis of the Scientific and Educational Center «Supercomputer Technologies». The reports cover all aspects of the use of supercomputers, parallel computing systems and distributed data processing methods for solving large computational problems. The seminar has a pronounced interdisciplinary nature, however, the various nuances of using supercomputer technologies are of interest to researchers from various fields.

PROGRAM

1. 16:30 - 17:15

On the role of non-blocking communications in the implementation of linear algebra algorithms on supercomputers

The effective use of asynchronous data exchange mechanisms is a prerequisite for the development of scalable linear algebra algorithms with sparse matrices. New features of the MPI-3 standard, in particular, the emergence of non-blocking collective operations, were the impetus for the development of modified mathematical algorithms aimed at blocking the time of global communications and computations. However, non-blocking operations are not always asynchronous in practice.

The report describes the developed methodology for testing the degree of asynchrony of non-blocking local and global communications. The importance of a correct understanding of the operation of non-blocking communications on the used computing equipment is demonstrated and examples of erroneous conclusions published in the literature about the properties of algorithms using asynchronous non-blocking global communications are discussed.

Links on the topic of work:

1) B. Krasnopolsky, Revisiting performance of BiCGStab methods for solving systems with multiple right-hand sides // Computers & Mathematics with Applications, 2019, https://doi.org/10.1016/j.camwa.2019.11.025

2) A. Medvedev. Towards benchmarking the asynchronous progress of non-blocking MPI point-to-point and collective operations // Proceedings of ParCo conference, 2020 (in press).

3) B. Krasnopolsky. Predicting Performance of Classical and Modified BiCGStab Iterative Methods // Proceedings of ParCo conference, 2020 (in press).

4) A. Medvedev. IMB-ASYNC benchmark. https://github.com/a-v-medvedev/mpi-benchmarks

2. 17:15 - 18:00

Investigation of the self-organization and physical properties of a platinum-copper surface alloy

Investigation of the self-organization and physical properties of a platinum-copper surface alloy
The work is devoted to computer modeling of the formation and physical properties of surface alloys of Pt-Cu. In the framework of this work, the interatomic interaction potentials of Pt-Cu and Pt-Pt were selected, and a numerical method for modeling the formation of surface Pt-Cu alloys was developed. The atomic mechanisms, dynamics, and growth conditions of finger-like outgrowths and fractal clusters in the surface Pt / Cu (111) alloy were studied. In the Pt / Cu (001) surface alloy, the order – disorder phase transition was studied with a change in the platinum concentration, the dynamics of the dissolution of platinum clusters in copper, and the effect of platinum atoms on the electromigration rate of vacancy islands. All results were obtained on Lomonosov and Lomonosov 2 supercomputers.

Links on the topic of work:

1) Dokukin, S. A.; Kolesnikov, S. V.; Saletsky, A. M. et al. Growth of the Pt/Cu(111) surface alloy: Self-learning kinetic Monte Carlo simulations. JOURNAL OF ALLOYS AND COMPOUNDS. 2018. V. 763. P. 719-727.

2) Dokukin, S. A.; Kolesnikov, S. V.; Saletsky, A. M. Efficient energy basin finding method for atomistic kinetic Monte Carlo models. COMPUTATIONAL MATERIALS SCIENCE. 2018. V. 155. P. 209-215.

3) Dokukin, S. A.; Kolesnikov, S., V; Saletsky, A. M. Dendritic growth of the Pt-Cu islands on Cu(111) surface: Self-learning kinetic Monte Carlo simulations. SURFACE SCIENCE. 2019. V. 689, UNSP 121464.

4) Dokukin, S. A.; Kolesnikov, S. V.; Saletsky, A. M. Diffusion-mediated processes in Pt/Cu(001) surface alloy. SURFACE SCIENCE. 2020. V. 692, 121515.