Interference-aware cloud scheduling architecture for dynamic latency-sensitive workloads
Computing systems continue to evolve to facilitate increased performance when processing workloads in large data centers. Virtualization technology enables multiple applications to be created and executed on a single physical computer, yielding various advantages, including rapid provisioning of res...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2022 |
| País: | Brasil |
| Institución: | Pontifícia Universidade Católica do Rio Grande do Sul (PUCRS) |
| Repositorio: | Biblioteca Digital de Teses e Dissertações da PUC_RS |
| Idioma: | inglés |
| OAI Identifier: | oai:tede2.pucrs.br:tede/10326 |
| Acceso en línea: | https://tede2.pucrs.br/tede2/handle/tede/10326 |
| Access Level: | acceso abierto |
| Palabra clave: | Interference-aware Scheduling Dynamic Latency-sensitive Workloads Machine Learning Resource Management Cloud Computing Simulation Escalonamento Ciente de Interferência Cargas de Trabalho Dinâmicas Sensíveis à Latência Aprendizado de Máquina Gerenciamento de Recursos Computação em Nuvem Simulação CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO |
| Sumario: | Computing systems continue to evolve to facilitate increased performance when processing workloads in large data centers. Virtualization technology enables multiple applications to be created and executed on a single physical computer, yielding various advantages, including rapid provisioning of resources and better utilization of hardware. Cloud computing providers have adopted this strategy to use their infrastructure more efficiently, reducing energy consumption. However, our research in this field has shown that multiple cloud services contending for shared resources are susceptible to crossapplication interference, which can lead to significant performance degradation and consequently an increase in the number of broken service level agreements (SLA). Nevertheless, state-of-the-art resource scheduling in virtualized environments still relies mainly on resource capacity, adopting heuristics such as bin-packing, thus overlooking this source of overhead. But in recent years interference-aware scheduling has gained traction, and applications are now being classified based on their interference level and the proposal of static cost models and policies for scheduling co-hosted cloud applications. Preliminary results in this area already show a considerable improvement in the reduction of broken SLAs, yet we strongly believe that there are still opportunities to improve in the areas of application classification and dynamic scheduling strategies. Therefore, this work’s primary goal is to study the behavior of cloud applications’ interference profiles over their entire life cycle, and their susceptibility to workload variations, looking for opportunities to improve resource sharing in virtualized environments with novel dynamic scheduling strategies. To this end, we explored some specific research questions related to the dynamic nature of the process, such as: How can applications be classified based on resource interference in real-time? When should classifications be executed? How many levels should be used? When should they be scheduled? What are the trade-offs with migration cost? To answer all of these questions, we created an interference-aware scheduling architecture that integrates the aforementioned topics to better manage dynamic latencysensitive workloads in virtualized environments. The contributions of this study are: (i) an analysis of the impact of workload variations in the interference profile of cloud applications; (ii) a precise and optimized way to classify applications in real-time; (iii) a novel dynamic interference-aware scheduling strategy for cloud applications; and (iv) a dynamic architecture that combines the above techniques to deliver efficient interference-aware scheduling in virtualized environments. Our results show an average 25% improvement of overall resource utilization efficiency with our architecture compared to related studies. |
|---|