A study on overcommitment in cloud providers

• Ano de defesa: 2025
• Autor(a) principal: Zhi, Jiang
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Alocação de recursos; Cloud computing; Computação em nuvem; Escalonamento; Overcommitment; Resource allocation; Scheduling; Sobrealocação
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/45/45134/tde-20082025-120108/

Cloud computing has become a key technology of the 21st century, widely recognized for its scalability, cost-effectiveness, and flexibility. Despite its advantages, resource inefficiencies remain a significant challenge for cloud service providers. For example, studies on Google Borg, a cluster management system, showed that only about 20\\% of computational capacity and 40\\% of memory are utilized, leaving substantial resources unused. These inefficiencies are often caused by users overestimating their resource needs and the complexities of managing workloads in unpredictable environments. This study focuses on overcommitment policies as a solution to improve resource utilization. Overcommitment involves allocating more virtual resources than the physical hardware can support, effectively leveraging underutilized capacity. By carefully managing quality of service (QoS), this technique can boost efficiency while maintaining performance. The research examines state-of-the-art overcommitment strategies, introduces a data collection tool to gather workload trace information from cloud service providers, and develops a simulator to evaluate these policies under diverse workloads. The goal of this thesis is to develop tools for studying overcommitment, aiming to minimize resource waste and enhance the performance of cloud computing environments.