Microservice Auto-Scaling Algorithm Based on Workload Prediction in Cloud-Edge Collaboration Environment
Zijun Peng,Bing Tang,Wei Xu,Qing Yang,Ehsanullah Hussaini,Yuqiang Xiao,Haiyan Li
DOI: https://doi.org/10.1109/ithings-greencom-cpscom-smartdata-cybermatics60724.2023.00112
2024-01-01
Abstract:Cloud computing centrally consolidates hardware and computing resources, offering efficient and cost-effective services. However, as cloud computing centers are predominantly built and operated in a fully centralized fashion, the increased distance between these centers and users can lead to a decline in service quality. Real-time interaction and high business continuity are crucial in scenarios like traffic monitoring, AR/VR applications, and the Internet of Things (IoT). Edge computing is better suited to meet the demands of such latency-sensitive business needs. By analyzing and processing massive data directly at edge computing nodes, which focus on network edge devices, reliance on transmission resources is reduced, consequently improving the overall quality and performance of services. Nevertheless, resource-constrained edge nodes require efficient utilization of available infrastructure capacity to ensure specific service level objectives (SLO) for applications. Therefore, this paper introduces XScale, a cloud-edge collaborative system that enables microservices to adaptively scale elastically. XScale applies a Bi-LSTM with an attention mechanism to forecast the workload of microservices. When combined with mechanisms designed to handle burst traffic and a cloud-edge collaborative load forwarding strategy, it achieves both adaptive elastic scaling and proactive load forwarding. Experimental results, obtained using real-world microservice workloads, indicate that the XScale system can significantly reduce SLO violations by 88%, increase resource utilization by 15%, and decrease average response time by 21% when compared to existing advanced reactive scaling methods.