SOFTWARE AVAILABILITY ENHANCEMENT IN PREEMPTIBLE INSTANCE KUBERNETES CLUSTER

• Main Author: Dewi, Rika
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/55851
• Institution: Institut Teknologi Bandung

From users perspective, cloud is often seen as an unlimited resource that can be used anytime and anywhere. In order to create the illusion as an unlimited resource, cloud service providers must always provide excess resources that exceed users demand. To increase the efficiency usage of their resources, some cloud service providers rent out their excess resources at lower prices with various limitations. Google Cloud Platform is one of the cloud service providers that rents out its excess resources called preemptible instances. Various limitations of preemptible instances cause a decrease in the availability of the software running on it. In this study, a tool called preemptible lifecycle scheduler is implemented to enhance availability on top of preemptible instances. This is done by scheduling the termination of the preemptible instance so that it occurs outside the application's peak hour range. Based on experiments, the use of preemptible instances in Kubernetes clusters can reduce infrastructure costs by up to 53.085%, but the software will experience a decrease in availability and no graceful shutdown period. By using the preemptible lifecycle scheduler tool, it is proven that it can increase the availability of the software system up to 0.629% during peak hours and increase the chance of graceful shutdown period by 37.1429% to 75% on software that is terminated by the preemptible lifecycle scheduler tool when scheduling the instance life cycle.