Development of SCIT Supercomputers Family Created at the V. M. Glushkov Institute of Cybernetics, NAS of Ukraine, in 200
- PDF / 84,565 Bytes
- 5 Pages / 594 x 792 pts Page_size
- 55 Downloads / 153 Views
SOFTWARE–HARDWARE SYSTEMS DEVELOPMENT OF SCIT SUPERCOMPUTERS FAMILY CREATED AT THE V. M. GLUSHKOV INSTITUTE OF CYBERNETICS, NAS OF UKRAINE, IN 2002–2017 A. Golovynskyi,1† I. Sergienko,1† V. Tulchinskyi,1‡ A. Malenko,1† O. Bandura,1† S. Gorenko,1† O. Roganova,1†† and O. Lavrikova1†
UDC 004.386
Abstract. The year 2017 marks the 15th anniversary of the SCIT supercomputer project, which allows us to summarize the results and draw conclusions. In this paper, we discuss the evolution of SCIT architecture and statistics of the supercomputing center for years 2002–2017. These data will be useful for computer cluster developers and researchers who design resource management algorithms for computing clusters. Keywords: supercomputer, cluster, parallel programming. INTRODUCTION Computing clusters of SCIT series (Supercomputer for Intelligent Technologies) are custom-made research systems that were constructed on the best technological basis at that time and embodied original domestic developments in system software and calculus mathematics libraries. In the paper, we will make a retrospective analysis of the development of SCIT architecture and arrange statistical data of clusters use. Many major supercomputer centers analyze users’ statistical data. For example, [1] presents such analysis for Kraken supercomputer at Oak Ridge Laboratory (USA). Despite the difference in core number and geographic location, this system and SCIT have similar structures of computing tasks in scientific fields and similar parameters of their distribution. EVOLUTION OF THE SCIT PROJECT ARCHITECTURE Development of the architecture of SCIT family clusters consists of five main stages shown in Table 1. The first research prototype of computing cluster SCIT-0 was created in 2002–2004. The main ideas of the project are described in [2, 3]. The cluster had four nodes on AMD Duron 800, interconnect 100 Mb Fast Ethernet, and 160 Gb data storage based on array RAID-1. The diskless principle, where all the nodes are clones loaded from one image of the operating system, was a distinctive feature of this cluster unlike the classical Beowulf architecture. This has facilitated administration and allowed fast and synchronous modifications in system and applied software. The architectural solution was tested on this 1
V. M. Glushkov Institute of Cybernetics, National Academy of Sciences of Ukraine, Kyiv, Ukraine, [email protected]; ‡[email protected]; ††[email protected]. Translated from Kibernetika i Sistemnyi Analiz, No. 4, July–August, 2017, pp. 124–129. Original article submitted December 19, 2016. †
600
1060-0396/17/5304-0600 ©2017 Springer Science+Business Media New York
TABLE 1. SCIT Clusters Performance
Year
Name
Number of CPU Cores
Performance in Linpack, Rpeak, Tflops
2002 2004 2005 2007 2012
SCIT-0 SCIT-1 SCIT-2 SCIÒ-3 SCIT-4
4 48 64 725 448
0.02 0.19 0.28 5.3 18
cluster: common disk storage for home tech users, which was simultaneously installed at all nodes of the cluster. These solutions varied negligibly in the next projects.
Data Loading...
