Approche prédictive de l'efficacité énergétique dans les Clouds Datacenters

Abstract : With the democratization of digital technologies, the construction of a globalized cyberspace insidiously transforms our lifestyle. Connect more than 4 billion people at high speed, requires the invention of new concept of service provision and trafic management that are capable to face the challenges. For that purpose, Cloud Computing have been set up to enable Datacenters to provide part or total IT components needed by companies for timely services delivering with performance that meets the requirements of their clients. Consequently, the proliferation of Datacenters around the world has brought to light the worrying question about the amount of energy needed for their function and the resulting difficulty for the humanity, whose current reserves are not extensible indefinitely. It was therefore necessary to develop techniques that reduce the power consumption of Datacenters by minimizing the energy losses orchestrated on servers where each wasted watt results in a chain effect on a substantial increase in the overall bill of Datacenters. Our work consisted first in making a review of the literature on the subject and then testing the ability of some prediction tools to improve the anticipation of the risks of energy loss caused by the misallocation of virtual equipment on servers. This study focused particularly on the ARMA tools and neural networks which in the literature have produced interesting results in related fields. After this step, it appeared to us that ARMA tools, although having less performance than neural networks in our context, runs faster and are best suited to be implemented in cloud computing environments. Thus, we used the results of this method to improve the decision-making process, notably for the proactive re-allocation of virtual equipment before it leads to under-consumption of resources on physical servers or over-consumption inducing breaches of SLAs. Based on our simulations, this approach enabled us to reduce energy consumption on a firm of 800 servers over a period of one day by more than 5Kwh. This gain could be significant when considering the enormous size of modern data centers and projected over a relatively long period of time. It would be even more interesting to deepen this research in order to generalize the integration of this predictive approach into existing techniques in order to significantly optimize the energy consumption within Datacenters while preserving performance and quality of service which are key requirements in the concept of Cloud Computing
Complete list of metadatas

Cited literature [41 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-01830891
Contributor : Abes Star <>
Submitted on : Thursday, July 5, 2018 - 2:19:06 PM
Last modification on : Saturday, December 21, 2019 - 3:43:52 AM
Long-term archiving on: Monday, October 1, 2018 - 5:11:08 PM

File

ThFrejusGBAGUIDI.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01830891, version 1

Collections

Citation

Fréjus A. Roméo Gbaguidi. Approche prédictive de l'efficacité énergétique dans les Clouds Datacenters. Réseau de neurones [cs.NE]. Conservatoire national des arts et metiers - CNAM; Université d'Abomey-Calavi (Bénin), 2017. Français. ⟨NNT : 2017CNAM1163⟩. ⟨tel-01830891⟩

Share

Metrics

Record views

369

Files downloads

493