Skip to Main content Skip to Navigation

Auditer l’énergie. Avant de déployer ses modèles : Vers des optimiseurs verts de requêtes analytiques

Abstract : In today’s world, we rely heavily on digital equipment for work, entertainment and social purposes. Therefore,we are obliged to seek all means to save the energy consumed by their hardware components, software, as well as the applications they use. Database management systems (DBMS) are becoming energy sinkholes due to the massive explosion of data they must collect, process and store. The query processor is one of the most energy intensive components of the DBMS. Its role is to efficiently process queries. Given the volume of data and the complexity of the analytical queries, the study of the energy efficiency of this component is undoubtedly becoming a crucial and urgent issue. This importance was widely underlined in the Claremont report and co-signed by around thirty researchers, experts and architects of the database world. This point is then echoed in Beckman’s report which places energy efficiency as a challenge to be addressed in the field of massive data. The majority of actual query optimizers are designed to minimize input-output (IO) operations and try to exploit main memory (RAM) as much as possible. As a result, they generally ignore the energy aspects. In this thesis, to optimize the energy consumed by a DBMS, we propose a software-oriented approach that we call Energy Audit - Before Deployment, allowing the design of less energy-consuming query processors. This approach consists of five main steps: (1) audit of query processor components in order to understand its operation and determine its energy-sensitive parameters. (2) Development of a state of the art on existing solutions with the dual purpose of (i) providing a roadmap for designers and students to understand energy efficiency issues in the database world and (ii) assisting in the development of energy models. (3) Query processors energy modeling, this modeling is achieved by defining mathematical cost models dedicated to estimate the energy consumption of the system during queries execution. (4) Using deep machine learning techniques to identify values of energy-sensitive parameters belonging to both hardware and software components. To apply our approach, we have chosen three open-source data processing systems, two of which are disk oriented:PostgreSQL (a row storage system) and MonetDB (a column storage system) and a main memory oriented Hyrise (a hybrid storage system). We then validate our approach using a series of experiments using data from several benchmarks (TPC-H, TPC-DS and SSB) and an apparatus to capture energy (Watts
Keywords : Green IT
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Friday, October 1, 2021 - 1:21:10 PM
Last modification on : Saturday, October 2, 2021 - 3:09:42 AM
Long-term archiving on: : Sunday, January 2, 2022 - 7:07:49 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03361382, version 1




Simon Pierre Dembele. Auditer l’énergie. Avant de déployer ses modèles : Vers des optimiseurs verts de requêtes analytiques. Autre [cs.OH]. ISAE-ENSMA Ecole Nationale Supérieure de Mécanique et d'Aérotechique - Poitiers, 2021. Français. ⟨NNT : 2021ESMA0009⟩. ⟨tel-03361382⟩



Record views


Files downloads