Efficient Resource Allocation for Multi-Robot Collaboration via Traffic-Aware Pod Autoscaling

Abstract

In edge computing, the ability to autoscale becomes imperative for container-based Internet of Things (IoT) applications to dynamically accommodate fluctuations in demands from IoT devices. Although Kubernetes offers horizontal pod autoscaling, which adjusts pod allocation based on node resource status, it needs more consideration for the varying resource demands across nodes in edge environments, resulting in suboptimal resource utilization. Moreover, the lack of support for vertical pod scaling in Kubernetes limits its ability to adjust pod sizes proactively according to computational needs. To address these limitations and enhance the quality of IoT services in edge computing, this paper introduces the Traffic-Aware Pod Scaler (TAPS), an augmentation for Kubernetes. TAPS leverages real-time trafficaware resource autoscaling to execute multidimensional upscaling and downscaling actions based on node network traffic statistics. Experimental results demonstrate that integrating TAPS with Kubernetes significantly improves IoT application performance and average response time by up to $100 \%$ compared to using the horizontal pod autoscaler alone. These findings underscore the critical importance of tailored resource scaling based on network traffic distribution to optimize the performance of IoT applications in edge computing environments.