Scholarlink

Current incremental Software-Defined Networking (SDN) migration strategies face two critical limitations: the inability to effectively represent dynamic network conditions and prolonged transition periods due to single-node sequential migration approaches. Existing approaches in dynamic network environments treat all traffic matrices clusters equally, ignoring their varying frequencies of occurrence and the presence of outliers, while single-node sequential migration strategies delay the realization of SDN benefits. This paper investigates the effectiveness of combining weighted traffic matrices with batch-oriented migration in a Deep Reinforcement Learning (DRL)-driven SDN migration framework. We propose a new framework integrating a multi-stage hybrid clustering method for weighted traffic matrices with a DRL model that enables batch migration decisions. Our framework demonstrates superior clustering performance through improved Dunn Index (0.0933 vs 0.0045 for GEANT, 0.1083 vs 0.0815 for Abilene) and Davies-Bouldin Index (1.2295 vs 1.3589 for GEANT, 1.6856 vs 1.7264 for Abilene) compared to standalone K-means, while achieving better weighted MLU performance during crucial early migration phases. Our research contributes to the development of practical SDN migration strategies that effectively represent dynamic network environments while enabling faster and more efficient transition through batch migration.

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

DRL-Driven Batch-Oriented SDN Migration with Weighted Traffic Matrix Clustering for Dynamic Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Kai Yuan Tan,
  • Saw Chin Tan

摘要

Current incremental Software-Defined Networking (SDN) migration strategies face two critical limitations: the inability to effectively represent dynamic network conditions and prolonged transition periods due to single-node sequential migration approaches. Existing approaches in dynamic network environments treat all traffic matrices clusters equally, ignoring their varying frequencies of occurrence and the presence of outliers, while single-node sequential migration strategies delay the realization of SDN benefits. This paper investigates the effectiveness of combining weighted traffic matrices with batch-oriented migration in a Deep Reinforcement Learning (DRL)-driven SDN migration framework. We propose a new framework integrating a multi-stage hybrid clustering method for weighted traffic matrices with a DRL model that enables batch migration decisions. Our framework demonstrates superior clustering performance through improved Dunn Index (0.0933 vs 0.0045 for GEANT, 0.1083 vs 0.0815 for Abilene) and Davies-Bouldin Index (1.2295 vs 1.3589 for GEANT, 1.6856 vs 1.7264 for Abilene) compared to standalone K-means, while achieving better weighted MLU performance during crucial early migration phases. Our research contributes to the development of practical SDN migration strategies that effectively represent dynamic network environments while enabling faster and more efficient transition through batch migration.