Geodesic
Automatic detection and description of supercomputer network infrastructure
Numerical methods and programming, Tome 15 (2014) no. 4, pp. 560-568.

Voir la notice de l'article provenant de la source Math-Net.Ru

The supercomputing system performance increases each year. In particular, this is because of increasing the number of computational nodes, making the memory subsystem and communication network more complex, etc., which causes the reduction of reliability and system's efficiency. As a result, the on-line control and efficient autonomous functioning of supercomputer complexes become more and more important. In order to solve this problem, Moscow University Research Computing Center is currently developing the Octotron system whose main objective is to provide the maximum safety and fullest usage of the hardware. The Octotron system uses a model of a computer system that should contain main supercomputer components and their interconnection. In particular, this model should contain a description of communication networks. Such a description could be significantly complex in many cases; therefore, the automation of this process is needed. In this paper we describe a programming tool being developed in order to detect Ethernet and Infiniband network topology in supercomputer systems. For detecting Ethernet network topology, this tool collects SNMP data from all available nodes and modifies them on the basis of the proposed rules to achieve more precise results. In the case of Infiniband network, it collects the necessary information from the subnet manager. The results of using this tool on the Lomonosov and Chebyshev supercomputers installed at Moscow University are discussed.
Keywords: Ethernet, Infiniband, supercomputers, parallel computing, supercomputer topology, communication networks, network topology detection, Ethernet, Infiniband
Mots-clés : SNMP protocol. 
@article{VMP_2014_15_4_a1,
     author = {Vad. V. Voevodin and K. S. Stefanov},
     title = {Automatic detection and description of supercomputer network infrastructure},
     journal = {Numerical methods and programming},
     pages = {560--568},
     publisher = {mathdoc},
     volume = {15},
     number = {4},
     year = {2014},
     language = {ru},
     url = {http://geodesic.mathdoc.fr/item/VMP_2014_15_4_a1/}
}
TY  - JOUR
AU  - Vad. V. Voevodin
AU  - K. S. Stefanov
TI  - Automatic detection and description of supercomputer network infrastructure
JO  - Numerical methods and programming
PY  - 2014
SP  - 560
EP  - 568
VL  - 15
IS  - 4
PB  - mathdoc
UR  - http://geodesic.mathdoc.fr/item/VMP_2014_15_4_a1/
LA  - ru
ID  - VMP_2014_15_4_a1
ER  - 
%0 Journal Article
%A Vad. V. Voevodin
%A K. S. Stefanov
%T Automatic detection and description of supercomputer network infrastructure
%J Numerical methods and programming
%D 2014
%P 560-568
%V 15
%N 4
%I mathdoc
%U http://geodesic.mathdoc.fr/item/VMP_2014_15_4_a1/
%G ru
%F VMP_2014_15_4_a1
Vad. V. Voevodin; K. S. Stefanov. Automatic detection and description of supercomputer network infrastructure. Numerical methods and programming, Tome 15 (2014) no. 4, pp. 560-568. http://geodesic.mathdoc.fr/item/VMP_2014_15_4_a1/