KubeHICE: Performance-aware Container Orchestration on Heterogeneous-ISA Architectures in Cloud-Edge Platforms

Abstract

To manage edge applications more conveniently, cloud providers extend the container and the container orchestrator from the cloud to the edge. The existing container orchestrators are designed for homogenous clusters at the cloud, while for nodes in cloud-edge computing environments, the hardware is commonly heterogeneous. Thus, current container orchestrators cannot automatically deal with heterogeneous-ISA. Besides, even if the nodes in a cloud-edge platform have the same ISA, their performance may be significantly different. The main reason lies in that the resource model of current container orchestrators is designed for homogenous clusters and assumes that all CPU cores of every node have the same performance, which will lead to low resource utilization and overload as far as heterogeneous clusters are concerned. In this paper, we present the design and implementation of KubeHICE, a performance-aware container orchestrator for heterogeneous-ISA architectures in cloud-edge platforms, which extends open source Kubernetes by AIM (Automatic Instruction Set Architecture Matching) and PAS (Performance-Aware Scheduling), two novel functional approaches that we propose. AIM is proposed for handling heterogeneous ISA, it can automatically find a node that is compatible with ISAs supported by the containerized application. PAS is designed for scheduling containers according to computing capability of cluster nodes. For PAS, we build a CPU resource allocation model to describe the computing capability differences of CPUs in a heterogeneous-ISA cluster. Based on this model, PAS is capable of estimating single-thread performance of a CPU through the workloads, which therefore enables KubeHICE to estimate the performance differences between containers that are assigned to heterogeneous nodes. Experimental results show that KubeHICE adds no additional overhead to container orchestration and it is effective in performance estimation and resource scheduling. We demonstrate the advantages of KubeHICE in several real-world scenarios, proving for example up to 40% increase in CPU utilization when eliminating heterogeneity. We believe that KubeHICE is not only technically beneficial to containerized applications in heterogeneous cloud-edge platforms, but also can a referable solution for Kubernetes community to support future IoT or edge server enabled container orchestration.