Skip to Main content Skip to Navigation
Theses

Performance et gestion de ressources dans un cloud multi-virtualisé

Abstract : Companies use cloud computing to lower entry and usage costs of using information technologies as a resource. The main feature of cloud computing that enables these lower costs is the pooling of resources with other users. Pooling of resources is based on virtualization. The principle is to sell physical, concrete resources from a data-center as virtual, abstract resources. The core concern for both cloud providers and clients is the efficiency, on various axes, of the virtualization solution. The former wish to serve as many clients as possible with the given physical resources of the data-center. It puts an emphasis on the capability of the virtualization solution to allocate resources with limited waste. Another target for optimization is the power usage of the data-center. Efficient management of resources has a great effect on it. As for clients of cloud computing, they seek guarantees on performance, including predictability. Indeed, the virtualization solution must provide them with the same performance level despite abstraction of physical resources into virtual resources. Buying cloud resources as cheap as possible is also a priority. There are two main virtualization solutions: virtual machines and containers. Both have their own sets of benefits and drawbacks, on the axes of resource management and performance. However containers can be nested inside of virtual machines, thus building a multi-virtualized cloud. How best to use both solutions in this new environment? This thesis explores issues that arise from combining both virtualization solutions. It proposes new systems to gain better performance, to improve resource management and to provide cheaper cloud services using multi-virtualization. First, it describes a novel workload consolidation method to further reduce power usage ; second, an allocation algorithm for resources of containers that fixes their intrinsic issue of performance predictability ; and third, two joint systems that optimize multi-virtualized networking to improve performance and resource utilization, and save money on cloud usage. Theses works play at every level of nested virtualization in order to move the technology of multi-virtualization forward.
Complete list of metadata

https://hal.archives-ouvertes.fr/tel-03139720
Contributor : Mathieu Bacou Connect in order to contact the contributor
Submitted on : Friday, February 12, 2021 - 11:35:46 AM
Last modification on : Thursday, June 10, 2021 - 3:07:27 AM
Long-term archiving on: : Thursday, May 13, 2021 - 6:38:43 PM

File

Bacou.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-03139720, version 1

Citation

Mathieu Bacou. Performance et gestion de ressources dans un cloud multi-virtualisé. Calcul parallèle, distribué et partagé [cs.DC]. Institut National Polytechnique de Toulouse, 2020. Français. ⟨tel-03139720⟩

Share

Metrics

Record views

123

Files downloads

180