Collaborative Edge-Cloud Computing for Efficient Resource Utilization
The computational requirements for running a wide variety of applications continue to grow. When designing a system to support these applications, there are several options to consider. One of them is edge computing. Edge computing reduces latency and enhances data privacy by bringing computing power closer to the data source. However, unlike cloud servers, edge devices often have limited available resources. By balancing resource usage between edge and cloud computing, the effectiveness and performance of applications can be significantly improved.
The importance of edge-cloud balance
Achieving this balance requires an understanding of how edge and cloud computing resources can work together to meet a variety of application requirements. The framework used to implement this edge-cloud architecture is Kubernetes. However, Kubernetes does not, by itself, include comprehensive mechanisms for distributing workloads evenly between edge and cloud infrastructures.
Evaluating edge-cloud collaboration with Kubernetes
Therefore, this study investigated how workloads can be moved between edge and cloud environments using Kubernetes, taking into account computational requirements, existing edge workloads, and other relevant criteria. This approach aimed to reduce latency while maintaining or improving application performance. The goal of this research was to explore a collaborative edge-cloud computing system that meets application requirements while efficiently utilizing resources.
Testing Kubernetes cluster configurations
Various Kubernetes cluster configurations were investigated, including edge-only, cloud-only, and combined edge-cloud configurations. To test these configurations, a demo application was deployed on the Kubernetes clusters. Performance and latency of the various configurations were analyzed using a load testing program.
Proposed edge-cloud autoscaler
Finally, an improved edge-cloud autoscaler setup for Kubernetes was proposed and its performance was compared to the previously evaluated configurations.
