Skip to Main content Skip to Navigation

Efficient Management of Resources in Heterogeneous Platforms

Clément Mommessin 1, 2 
2 DATAMOVE - Data Aware Large Scale Computing
Inria Grenoble - Rhône-Alpes, LIG - Laboratoire d'Informatique de Grenoble
Abstract : The world of Information Technology (IT) is in constant evolution.With the explosion of the number of digital and connected devices in our everyday life, the IT infrastructures have to face an ever growing amount of users, computing requests and data generated.The Internet of Things have seen the development of computing platforms at the edge of the network to bridge the gap between the connected devices and the Cloud, called the Edge Computing.In the domain of High Performance Computing, the parallel programs executed on the platforms requires always more computing power in a search for improved performances.Besides, we observed in the past years a diversification of the hardware composing these infrastructures.This complexification of the (network of) computing platforms pose several optimisation challenges that can appear at different levels.In particular, it led to a need for better management systems to make an efficient usage of the heterogeneous resources composing these platforms.The work presented in this thesis focuses on resources optimisation problems for distributed and parallel platforms of the Edge Computing and High Performance Computing domains.In both cases, we study the modelling of the problems and propose methods and algorithms to optimise the resources management for better performances, in terms of quality of the solutions.The problems are studied from both theoretical and practical perspectives.More specifically, we study the resources management problems at multiple levels of the Qarnot Computing platform, an Edge Computing production platform mostly composed of computing resources deployed in heaters of smart-buildings.In this regard, we propose extensions to the Batsim simulator to enable the simulation of Edge Computing platforms and ease the design, development and comparison of data and jobs placement policies in such platforms.Then, we design a new temperature prediction method for smart-buildings and propose a formulation of a new scheduling problem with two-agents on multiple machines.In parallel, we study the problem of scheduling applications on hybrid multi-core machines in the objective of minimising the completion time of the overall application.We survey existing algorithms providing performance guarantees on the constructed schedules and propose two new algorithms for different settings of the problem, proving performance guarantees for both.Then, we conduct an experimental campaign to compare in practice the relative performance of the new algorithms with existing solutions in the literature.
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Wednesday, March 24, 2021 - 10:32:09 AM
Last modification on : Wednesday, July 6, 2022 - 4:23:00 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03179102, version 1


Clément Mommessin. Efficient Management of Resources in Heterogeneous Platforms. Performance [cs.PF]. Université Grenoble Alpes [2020-..], 2020. English. ⟨NNT : 2020GRALM065⟩. ⟨tel-03179102⟩



Record views


Files downloads