Optimized Placement of Service Function Chains in Edge Cloud with LSTM and ILP

Edge computing has emerged as a transformative approach for reducing latency and enhancing network performance by placing computing resources closer to data sources and end users via edge nodes. This approach addresses the delays inherent in traditional cloud computing architectures, where data must travel long distances to centralized data centers. Within this architecture, Network Function Virtualization (NFV) enables network service deployment by executing Virtual Network Functions (VNFs) on virtualized infrastructure, establishing Service Function Chains (SFCs) to manage network traffic efficiently. In this work, we address the challenge of optimizing SFC placement that combines Long Short-Term Memory (LSTM) networks and Integer Linear Programming (ILP). The LSTM model predicts dynamic traffic patterns and resource demands of edge nodes, allowing for a more accurate anticipation of future network conditions. These predictions are then used to formulate an ILP-based optimization model that determines the optimal placement of SFCs to minimize deployment costs while considering constraints like delay, memory, CPU, and bandwidth. We implemented and evaluated the proposed approach in a real-time KubeEdge cluster consisting of one master node and eight edge nodes. The results demonstrate that the LSTM-ILP scheduler significantly improves SFC placement efficiency by effectively identifying underutilized nodes, leading to reduced response times and optimized resource usage. The LSTM-ILP scheduler improves response time by 12.5% compared to the default Kubernetes scheduler when handling 50,000 requests.