Scholarlink

<p>In a Virtual Function Network, Service Function Chains (SFCs) must be migrated in response to network status changes and fluctuating resource demands. To minimize the negative impact on service quality during such migrations, this paper presents a novel live SFC migration approach named VNF-SFCM-DPSTR that ensures a seamless transition—where the original SFC remains active until the new SFC is fully established. The proposed approach begins by constructing an SFC migration cost model that considers factors such as bandwidth, CPU, and memory consumption. Subsequently, an online training mechanism based on Bidirectional Gated Recurrent Units is developed to predict resource demand in the substrate network. This enables the accurate identification of potentially overloaded links and nodes, allowing sufficient time to complete the migration process. For nodes predicted to be overloaded, the Virtual Network Function (VNF) with the least impact on the SFC is selected for replication. A new host node for this VNF is then determined using an enhanced Discrete Particle Swarm Optimization algorithm. Finally, a dynamic path segmentation algorithm is employed to reroute traffic among the new host nodes containing the replicated VNFs, thereby constructing the migrated SFC. Once the new chain is fully operational, the original SFC is decommissioned. Simulation results demonstrate the proposed approach’s superiority over TPGDM-DBN and RATO-DSFCM in terms of prediction accuracy, model efficiency, migration time, and overall SFC reliability. The implementation of our proposed method is now publicly accessible at <a href="https://github.com/LZ825/VNF-SFCM-DPSTR.">https://github.com/LZ825/VNF-SFCM-DPSTR.</a></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Service Function Chain Live Migration with Bi-GRU Resource Prediction in Substrate Networks

  • 期刊论文

  • 发表日期：

    2026年2月12日

查看全文
  • Zhen Liu,
  • Zhiping Zhang,
  • Vladimir Ciric,
  • Changda Wang

摘要

In a Virtual Function Network, Service Function Chains (SFCs) must be migrated in response to network status changes and fluctuating resource demands. To minimize the negative impact on service quality during such migrations, this paper presents a novel live SFC migration approach named VNF-SFCM-DPSTR that ensures a seamless transition—where the original SFC remains active until the new SFC is fully established. The proposed approach begins by constructing an SFC migration cost model that considers factors such as bandwidth, CPU, and memory consumption. Subsequently, an online training mechanism based on Bidirectional Gated Recurrent Units is developed to predict resource demand in the substrate network. This enables the accurate identification of potentially overloaded links and nodes, allowing sufficient time to complete the migration process. For nodes predicted to be overloaded, the Virtual Network Function (VNF) with the least impact on the SFC is selected for replication. A new host node for this VNF is then determined using an enhanced Discrete Particle Swarm Optimization algorithm. Finally, a dynamic path segmentation algorithm is employed to reroute traffic among the new host nodes containing the replicated VNFs, thereby constructing the migrated SFC. Once the new chain is fully operational, the original SFC is decommissioned. Simulation results demonstrate the proposed approach’s superiority over TPGDM-DBN and RATO-DSFCM in terms of prediction accuracy, model efficiency, migration time, and overall SFC reliability. The implementation of our proposed method is now publicly accessible at https://github.com/LZ825/VNF-SFCM-DPSTR.