基于Gossip机制的自组织覆盖网络拓扑管理协议(英文)

• 作 者: 陈铙 胡瑞敏 朱永琼
• 机 构: National Engineering Research Center for Multimedia Software,Wuhan University,
• 出 处: 中国通信
• 发表日期: 2011年05期
• 关键词: overlay balancing topology dissemination neighbor 网络拓扑 organizing simplicity qui

Gossip-based protocols have attracted more and more attention because of their simplicity and reliability.They can be applied to large-scale overlays for solving problems such as topology management,information dissemination,and aggregation.However,previous works sample nodes by their indegrees,without considering the differences in capability among nodes,and result in losing global load balancing.This paper proposes a load balancing gossip protocol for self-organizing overlays-LBTMP(Load-Balancing Topology Management Protocol),which takes into account the differences in capability among nodes and real loads.The novel protocol takes remainder service ability as the determinant for node selection metric,making light loading nodes from local neighbor view as returned samples preferentially.In the meantime,LBTMP selects light loading nodes preferentially for topology information exchange,which can diffuse light loading nodes over the whole overlay more quickly.Simulations show that returned sample node selection is biased to light loading nodes in a global view,and the overlay tends to load balancing.

Gossip-based protocols have attracted more and more attention because of their simplicity and reliability. They can be applied to large-scale overlays for solving problems such as topology management, information dissemination, and aggregation. However, previous works sample nodes by their indegrees, without considering the differences in capability among nodes, and result in losing global load balancing. This paper proposes a load balancing gossip protocol for self-organizing overlays-LBTMP (Load-Balancing Topology Management Protocol) nodes and real loads.The novel protocol takes the remainder service ability as the determinant for node selection metric, making light loading nodes from local neighbor view as returned samples preferentially. In the meantime, LBTMP selects light loading nodes preferentially for topology information exchange, which can diffuse light loading nodes over the whole overlay more quickly.Simulations show that return ed sample node selection is biased to light loading nodes in a global view, and the overlay tends to load balancing.