Dynamic Service Function Chain Placement in Edge-Computing Environments Using LSTM and Auction Based Heuristics
Year: 2026
Type: conference-paper
Abstract: The placement of Service Function Chains (SFCs) in edge computing settings becomes increasingly difficult especially when workloads change constantly and hardware resources are limited, more so when applications need minimum latency. Most of the existing SFC placement approaches heavily depend on reactive scheduling or static rules, which are not always effective in such highly dynamic situations. We have developed a framework for adaptive SFC placement that combines decision, based scheduling with workload prediction for a cloud, edge Kubernetes setup. Based on the CPU, memory, and round, trip time (RTT) data collected in the past from the cluster, an LSTM model identifies temporal patterns. Using sliding windows of previous measurements, the worker nodes are categorized as lightly loaded, moderately loaded, and heavily loaded. The model's classification accuracy on the testbed dataset is 96.34%; on the Alibaba Cluster Trace dataset, it is 92.22%. By not relying solely on the present states of the system, these forecasts enable the scheduler to make placement decisions based on expected future states. An auction, based heuristic method is applied to select a suitable node while encouraging stability and fairness among the nodes when they happen to belong to the same load category. The suggested technique is tested under several traffic scenarios and compared with an LSTM only approach and the default Kubernetes scheduler. Experimental results show that the hybrid LSTM and auction, based method can increase throughput by around 22-25% and at the same time have a lower average latency. In general, these results suggest that the issue of SFC placement in dynamic edge computing environments can be more effectively addressed by integrating predictive modeling with fairness, aware decision making.
Institution: KLE Technological University
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204802939