Distributed Scheduling with ATLAS

In cloud-based architectures, effective resource management and job distribution are indispensable, with schedulers like Kubernetes playing a pivotal role. While auto-scaling, the provision or removal of cluster resources based on current workload, has become a standard solution for meeting Service Level Agreements (SLAs), this thesis aims to study if real-time scheduling algorithms such as EDF (Earliest Deadline First) and LLF (Least Laxity First), together with the execution time prediction feature from ATLAS (Auto-Training Look-Ahead Scheduler) could improve SLAs without auto-scaling. In this talk, I will delve into the design and implementation of EDF and LLF schedulers within the Kubernetes framework, presenting a comparative analysis of the scheduling outcomes between the new and default schedulers.

(Master Thesis Defense)