Container Orchestration in Heterogeneous Edge Computing Environments

The emerging latency-sensitive applications and Internet of Things technology have resulted in the development of Edge computing. Therefore, improving Quality-of-Service (QoS) requirements such as response time is a fundamental goal in Edge environments. However, as edge devices are heterogeneous and resource-constrained, placing replicas of software containers in these environments is not a trivial task. Yet, the scheduler of well-known container orchestration tools such as Kubernetes places pods in nodes by only considering available resources (CPU, memory, etc.). Our study aims to minimize application latency by optimizing resource allocation through efficient scheduling. We customize the Kubernetes scheduler to assign pods to nodes with the least response time and integrate it with a customized autoscaler to scale up/down replicas. We evaluate our algorithm on a small-scale cluster with two types of deployments (web and object detection services), and evaluation results show better response time and throughput than the default scheduler.