Research on multi-level hybrid resource allocation fairness model based on kubernetes

• 1、School of computer science, Beijing University of Posts and Telecommunications,Beijing 100876

Abstract：Kubernetes is now the mainstream container orchestration engine in the industry. In the multi- tenant batch task processing scenario, Volcano, the mainstream scheduler, adopts a multi-level DRF algorithm to achieve fair resource allocation by balancing the dominant resource share of multiple tenants and the dominant resource share between multiple jobs of the same user.However, DRF algorithm only considers the resource usage of each tenant at the allocation time, and does not consider the fairness of resource allocation from the time dimension, which may lead to unfair resource allocation.H-MRF algorithm combines LT-DRF algorithm and LT-AF algorithm to achieve long-term resource allocation fairness, but the algorithm fails to consider the complexity of multi tenant tasks.Combined with H-MRF algorithm and multi-level DRF algorithm, this paper proposes a multi-level hybrid resource allocation fairness (MHRAF) model to realize the long-term resource allocation fairness among multiple tenants, the priority queuing in the complex situation of multi job and multi task of the same tenant, and the DRF characteristics between multiple jobs.The scheduler MyVolcano is implemented based on the MHRAF model, and some comparative experimentsare carried out in kubernetes cluster.Experimental results show that MHRAF algorithm can achieve long-term fairness of multi-tenant resource allocation and more reasonable task resource allocation.

Keywords： Cloud Computing Kubernetes Resource Allocation Fairness Multi-tenant Long-Term