An Empirical Design and Implementation of Job Scheduling Enhancement for Kubernetes Clusters

Batch job scheduling is a popular method in scheduling topics achieving considerable results. However, it is novel to bring those achievements to the cloud where problems of environments would limit algorithms. Although some projects have been introduced recently with benefits directly targeting machine learning jobs in cloud-native environments, there are gaps to be fulfilled. Consequently, this paper proposes an empirical design and implementation of deadline-aware enhancement of job scheduling for cloud-native environments. The proposal’s target is to automatically monitor, provision, and maintain machine learning jobs for batch job scheduling in cloud-native environments. The performance evaluation shows that the proposal has succeeded in reducing the average response time of machine learning jobs scheduled by different algorithms.