Mitigating Auto-scaling Delays in Elastic-Docker for Better Responsiveness of Burst and Dynamic Workloads

Dynamic assignment of resources to containers at runtime is crucial for improving users’ quality of experience while maximizing resource utilization in a cloud environment. ElasticDocker is a threshold-based system for vertically auto-scaling Docker containers in the cloud. This paper presents a set of improvements aimed at improving the responsiveness of ElasticDocker to burst workloads. The objective involves generating predictions and assessing resource utilization for the original ElasticDocker implementation and the proposed solution. The analysis aims to provide valuable insights into resource utilization patterns and optimization opportunities. We compared both systems in the context of a video conversion workload. The economic feasibility and benefits of adopting the proposed improvements can be determined by comparing resource utilization and cost implications. The results show that the proposed improvements improve responsiveness in burst workloads allowing the system to finish the transcoding job 33.6% faster while consuming 6.2% less memory. We also investigate the impact of ElasticDocker with the proposed improvements on the performance and compare the results with the original ElasticDocker. The results show that when the number of requests increases and requires more resources, ElasticDocker with the proposed improvements reacts to provision resources accordingly, therefore the response time decreases, and the performance increases. The proposed set of improvements is simple yet effective and provides useful insights and a cornerstone in terms of how to start applying the elasticity of Docker containers for building a complete monitoring and adjusting system for running containers for researchers and practitioners.