Dynamic provisioning of container registries in edge computing infrastructures

• Ano de defesa: 2024
• Autor(a) principal: Araujo, Lucas Roges de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul; Escola Politécnica; Brasil; PUCRS; Programa de Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Container Registry; Edge Computing; Docker; Registro de Contêiner; Computação de Borda; CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO
• Link de acesso: https://tede2.pucrs.br/tede2/handle/tede/11652

The proliferation of mobile devices and sensors has provoked the development of latency-sensitive and resource-intensive applications. Although devices and sensors that generate data might have limited processing and storage capacity, cloud computing infrastructures offer the necessary scalability to process these demands. However, the distance between end users and cloud data centers introduces communication overhead. In this sense, edge computing emerges to extend cloud infrastructures to locations near the end user. The inherited proximity of the paradigm improves latency and bandwidth for applications with strict demands. Additionally, the association with container-based virtualization improves Quality of Service (QoS) and Quality of Experience (QoE). The lightweight aspect of this technology provides faster application provisioning and lower resource overhead. However, provisioning container-based applications suffers from a significant overhead while downloading content from container registries. Although several authors propose using distributed registries to tackle this problem, their techniques focus exclusively on provisioning time, often neglecting key metrics for edge computing infrastructures, such as latency or resource usage. In addition, other authors adopt complex techniques, such as registry migration, which can affect network utilization. In response to this scenario, we propose dynamic provisioning of container registries with two strategies (LMDyn and MODyn) that allocate container registries to edge servers based on infrastructure analysis without requiring resource migration. Our evaluation demonstrates significant reductions in resource usage with LMDyn and a balanced trade-off between provisioning time, latency, and resource usage with MODyn