CC BY
53. Boris Glinskiy, Anna Sapetina, Valeriy Martynov, Dmitry Weins, Igor Chernykh The Hybrid-Cluster Multilevel
Approach to Solving the Elastic Wave Propagation Problem. // Communications in Computer and Information Science
book series, Springer, (CCIS, V. 753), Pages 261-274.
4. B.M. Glinskiy, G.F. Zhernyak, P. A. Titov. System of intelligent support for solving geophysical problems. Proc. of the
6th Russian Supercomputing Days-2020, Moscow, Sept. 21�22, 2020. P. 11�18. DOI: 10.29003/m1406.RussianSCDays-2020.
Thermal efficiency and computing productivity metrics for data center operations
A. A. Grishina1*, M. Chinnici2, A.-L. Kor3, D. De Chiara4, J.-P. Georges5, E. Rondeau5
1Simula Research Laboratory, Oslo, Norway
2ENEA-R.C. Casaccia, Rome, Italy
3Leeds Beckett University, Leeds, UK
4ENEA-R.C. Portici, Portici (Naples), Italy
5CRAN-CNRS, University of Lorraine, Nancy, France
Email: a.a.grishina17@gmail.com
DOI 10.24412/cl-35065-2021-1-02-93
Data Centers (DCs) provide scalable on-demand computing and networking services to a growing number
of businesses, government, health organizations, and end-users. Naturally, DCs� power demand for cooling and
IT equipment has increased over the last few years [1]. However, some energy utilized in DCs is wasted for
both inefficient cooling and applications that do not run properly on computing clusters [2]. We present a
framework for the investigation of (a) hidden thermal pitfalls using thermal metrics [3] and guidelines [4] as
well as machine learning for assessment of thermal factors affecting individual servers [5] and (b) energy
waste that is caused by ineffective computations and can be translated into carbon waste evaluation with the
help of a proposed metric [6]. The framework has been tested on job scheduling reports and (a) thermal me-
ters of CRESCO4 and CRESCO6 clusters in ENEA Portici DC. The work results in a set of recommendations on
how the productivity assessment could drive a new power and thermal efficiency management strategy.
This research work has been supported and funded by the PERCCOM Erasmus Mundus Program of the European Un-
ion [7].
*The work was conducted while the author followed EMJMD PERCCOM [7] at ENEA Casaccia R.C., Rome, Italy
References
1. A. Shehabi, et al., �United States Data Center Energy Usage Report,� Lawrence Berkeley Natl. Lab. Berkeley, CA,
Tech. Rep., pp. 166, 2016.
2. Chinnici, M.; Capozzoli, A.; Serale, G. Measuring energy e_ciency in data centers. In Pervasive Computing: Next
Generation Platforms for Intelligent Data Collection; Dobre, C., Xhafa, F., Eds.; Morgan Kaufmann: Burlington, MA, USA,
2016; Chapter 10; pp. 299�351. ISBN 9780128037027.
3. Capozzoli, A.; Serale, G.; Liuzzo, L.; Chinnici, M. Thermal metrics for data centers: A critical review. Energy Procedia
2014, 62, 391�400.
4. ASHRAE Technical Committee 9.9, �Thermal Guidelines for Data Processing Environments � Expanded Data Center
Classes and Usage Guidance,� 2011.
5. Grishina A, Chinnici M, Kor A-L, Rondeau E, Georges J-P. A Machine Learning Solution for Data Center Thermal
Characteristics Analysis. Energies. 2020; 13(17):4378. https://doi.org/10.3390/en13174378
6. Grishina, A.; Chinnici, M.; De Chiara, D.; Guarnieri, G.; Kor, A.-L.; Rondeau, E.; Georges, J.-P. DC Energy Data Meas-
urement and Analysis for Productivity and Waste Energy Assessment. In Proceedings of the 2018 IEEE International Con-
ference on Computational Science and Engineering (CSE), Bucharest, Romania, 29�31 October 2018; IEEE: Piscataway, NJ,
USA, 2018; pp. 1�11, ISBN 978-1-5386-7649-3.
7. A. Klimova, E. Rondeau, K. Andersson, J. Porras, A. Rybin, and A. Zaslavsky,�An international Master�s program in
green ICT as a contribution to sustainable development, J. of Cleaner Production, vol. 135, pp. 223�239, 2016.
Simulations of viscous incompressible fluid flows on grids with unmatched interfaces in the LOGOS software
package
A. V. Korotkov, S. V. Lashkin, A. S. Kozelkov
FSUE �Russian Federal Nuclear Center � All-Russian Research Institute of Experimental Physics�, Sarov, Nizhny
Novgorod Region
Email: alvladkor79@mail.ru
DOI 10.24412/cl-35065-2021-1-01-73
At present, numerical simulations of industry-specific hydrodynamic and aerodynamic problems are based
on solving three-dimensional Navier-Stokes equations on arbitrary unstructured grids [1, 2]. Dividing complex
initial geometries into simpler fragments makes it easier to construct grid models and yields higher-quality
grids. Such grid models are usually composed of unmatched grid fragments. CFD simulations on this kind of
grids require special unmatched grid interfaces to be developed.
This paper describes a numerical method, which considers specific aspects of solving the Navier-Stokes
equations in viscous incompressible flow simulations in the vicinity of interfaces between unmatched arbitrary
unstructured grid fragments. The numerical method presented in this paper is implemented based on the Rus-
sian LOGOC software package [3]. Performance of this method is demonstrated by three-dimensional simula-
tions of a turbulent flow in a circular diffuser.
References
1. Pogosyan M.A., Savelievskikh E.P., Shagaliev R.M., Kozelkov A.S., Strelets D.Yu., Ryabov A.A., Kornev A.V., Deryugin
Yu.N., Spiridonov V.F., Tsiberev K.V. Application of Russian supercomputer technologies to develop the advanced models
of aviation technology // J. VANT, Ser. Mathematical Modeling of Physical Processes, 2013, issue 3, p. 3-17. [ In Russian].
2. Betelin V.B., Shagaliev R.M., Aksenov S.V., Belyakov I.M., Deryuguin Yu.N., Kozelkov A.S., Korchazhkin D.A., Nikitin
V.F., Sarazov A.V., Zelenskiy D.K. Mathematical simulation of hydrogen�oxygen combustion in rocket engines using
LOGOS code // Acta Astronautica 2014, v. 96, p.53�64.
3. Kozelkov A.S., Shagaliev R.M., Kurulin V.V., Yalozo A.V., Lashkin S.V. Analysis of supercomputer potential as applied
to hydrodynamic scalable numerical simulations in industrial applications // Computational Mathematics and
Mathematical Physics. 2016. V. 56. Iss. 8. P. 1524-1535. [In Russian].
Iterative refinement approach for improving convergence of the mixed precision linear solvers
B. I. Krasnopolsky
Institute of Mechanics, Lomonosov Moscow State University
Email: krasnopolsky@imec.msu.ru
DOI 10.24412/cl-35065-2021-1-01-74
The use of the lower precision floating point calculations is an evident way to speed up the solution of sys-
tems of linear algebraic equations. The widely used approach for iterative methods assumes the use of the
mixed precision calculations when the preconditioner is performed with the lower precision. In practice it can
provide 10-15 % calculations speedup. The more attractive way is the use of the lower precision for the whole
solver, which can reduce the calculation time by a factor of 1.6-1.8, but may affect the convergence rate.
The current talk discusses an alternative variant of introducing the mixed precision calculations, which
provides the double precision solution accuracy, but allows to perform most of the calculations in the single
precision [1]. The algorithm is based on the iterative refinement procedure and combines two nested iterative
