Abstract. 

Communication networks play a key role in parallel high-performance computers, called supercomputers, in terms of their performance and cost. Efficiency and cost of communication networks mostly depend on the network topology. In this paper, we describe the MKNS topology developed for communication networks of exascale parallel computers. We provide a comparative analysis of the basic communication characteristics of the MKNS, 4D-Torus, 6D-Torus and Fat Tree topologies, which are used to build communication networks for parallel high-performance computers.

Keywords: 

network, high-performance computer, MKNS topology

PP. 15-26.

DOI 10.14357/20718632210302

 

References

1. Stepanenko S. A. 2016. Multiprotsessornye sredy super- EVM. Masshtabirovanie effektivnosti. [Multiprocessor supercomputer environments. Efficiency scaling] Moscow: Fizmatlit. 312 p.

2. InfiniBand Trade Assosiation, InfiniBand Architecture Specification Volume 1 – Release 1.2.1. Available at: https://www.infiniband.com (accessed November 10, 2019).

3. T. P. Morgan, How Cray Makes Ethernet Suited for HPC and AI with Slingshot. Available at: https://nextplatform/2019/08/16/(accessed December 6, 2019).

4. D. Chen, et al. 2012. The IBM Blue Gene/Q Interconnection Network and Message Unit. In Proceedings of the International Conference on High Performance Computing, Networking, Storage and Analysis (SC):1-12.

5. Y. Ajima, T. Inoue, S. Hiramoto, and T. Shimizu. 2012. Tofu: Interconnect for the K computer. Fujitsu Scientific and Technical Journal vol. 48, no.3:280-285.

6. Simonov A. S., Makagon D. V., Gabin I. A., Shcherbak A. N., Syromyatnikov E. L., Polyakov D. A. 2014. Pervoe pokolenie vysokoskorostnoy kommunikatsionnoy seti «Angara» [The first generation of the “Angara” highspeed communication network.] Science technology. V. 15, 1: 21-28.

7. Kholostov A. A. 2013. Masshtabiruemaya sistema mezhprotsessornykh obmenov 10G. [10G scalable interconnection system]. Trudy 2-go natsionalnogo superkompyuternogo foruma. [Proceedings of the Second National Supercomputing Forum.] Pereslavl-Zalessky. 126 -134.

8. Top500 list. Available at: http://www.top500.org (accessed February19, 2021).

9. Stepin A. Rossiskiy interkonnekt «Angara-2» [Russian interconnect “Angara-2”]. Available at:

https://servernews.ru/1033236 (accessed February 11, 2021).

10. Dongarra J. 2016. Report on the Sunway TaihuLight System. University of Tennessee Department of Electrical Engineering and Computer Science Tech. Report UTEECS- 16–742.

11. Kim J., Dally W. J., Scott S., Abts D. 2008. Technology-Driven, Highly-Scalable Dragonfly Topology. Computer Architecture 08:77 – 88.

12. Cray XC30 System. Available at: https://www.nersc.gov (accessed February 01, 2021).

13. Roberto Peñaranda, Crispín Gómez, María E. Gómez, Pedro López & Jose Duato 2016. The K-ary N-direct Sindirect family of topologies for large-scale interconnection network. The Journal of Supercomputing volume 72:1035–1062, Available at:

http://doi.org/10.1007/s11227-016-1640-z) (accessed November 02, 2019).

14. Basalov V. G., Kholostov A. A. 2016. Perspektivnaya gibridnaya topologiya KNS dlya system mezhprotsessornykh obmenov na baze SMPO-10G. [A promising hybrid topology KNS for interconnection systems based on SMPO-10G] Voprosy Atomnoy Nauki i Tekhniki. Matematicheskoe modelirovanie fizicheskikh protsessov. [Mathematical Modeling of Physical Processes] v.3:62-69.