Scholarlink

<p>At the present stage, despite that the SDN network size changes frequently and traffic is dynamic, efficient routing in large-scale software defined networking (SDN) networks seems to be one of toughest challenges, thus it is extremely urgent to design more flexible and intelligent optimization strategies. The objective of this writing is that we serve a balance between the present loose global coordination more in line with metaheuristic search, and learning-based fine-tuning, which limits routing flexibility (on one hand) and multi-objective outcome. A hybrid approach is utilized by combining a derivative African Vulture Routing Optimization (AVRO)-algorithm enhanced with edge- betweenness-based initialization, adaptive-gain control, and Lévy flight–based exploration—with a Deep Q-Network (DQN) fine-tuning stage over capacity-aware k-shortest-path domain featuring novelty through multipath allocation of flow, multiobjective cost estimation and reinforcement-learning-enabled link-weight update. The method is able to achieve far higher fitness, throughput and gain/loss ratio across the GBN, GEANT2, and NSFNET topologies showing consistent convergence stability and routing efficiency over classical and learning-based competitors. This result suggests the joint utilization of evolutionary intelligence and deep reinforcement learning improves scalability, response time, and optimization quality in programmable networks. Our results are relevant in this context, since they provide sound constitutive grounds for subsequent self-optimizing SDN routing models.</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

Hybrid AVRO–DQN Optimization for Improved Routing Performance in Software-Defined Networks

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Anishetty Laxmi Prasanna,
  • G.S. Raghavendra

摘要

At the present stage, despite that the SDN network size changes frequently and traffic is dynamic, efficient routing in large-scale software defined networking (SDN) networks seems to be one of toughest challenges, thus it is extremely urgent to design more flexible and intelligent optimization strategies. The objective of this writing is that we serve a balance between the present loose global coordination more in line with metaheuristic search, and learning-based fine-tuning, which limits routing flexibility (on one hand) and multi-objective outcome. A hybrid approach is utilized by combining a derivative African Vulture Routing Optimization (AVRO)-algorithm enhanced with edge- betweenness-based initialization, adaptive-gain control, and Lévy flight–based exploration—with a Deep Q-Network (DQN) fine-tuning stage over capacity-aware k-shortest-path domain featuring novelty through multipath allocation of flow, multiobjective cost estimation and reinforcement-learning-enabled link-weight update. The method is able to achieve far higher fitness, throughput and gain/loss ratio across the GBN, GEANT2, and NSFNET topologies showing consistent convergence stability and routing efficiency over classical and learning-based competitors. This result suggests the joint utilization of evolutionary intelligence and deep reinforcement learning improves scalability, response time, and optimization quality in programmable networks. Our results are relevant in this context, since they provide sound constitutive grounds for subsequent self-optimizing SDN routing models.