延迟容忍网络中一种基于概率接纳和丢弃的拥塞控制算法

发布时间：2018-12-31 09:51

【摘要】：链路的间歇性连通以及稳定的端到端路径的缺乏使得延迟容忍网络(delay tolerant network)中经常采用"存储-携带-转发"的方式来保证消息传输的到达率。然而由于网络资源受限,该转发方式下产生的大量的消息副本将造成巨大的资源消耗,最终导致网络拥塞。提出一种基于概率接纳和丢弃(probabilistic acceptance and drop,PAD)的拥塞控制算法PAD。该算法结合了队列长度和输入/输出速率来检测拥塞,各个节点根据当前的拥塞状态来确定接收和丢弃消息的概率,从而实现较小的开销和较高的消息到达率。此外,基于生灭模型构造了消息副本数的连续时间马尔可夫链,并对消息到达率进行了理论分析。理论分析和仿真结果证明,与其他算法相比,PAD算法在保证较小的网络开销和较短的端到端延迟的同时,消息到达率显著地提高了130%以上。
[Abstract]:The intermittent connectivity of links and the lack of stable end-to-end paths lead to the "storage-port-to-forward" approach to ensure the arrival rate of messages in delay tolerance networks (delay tolerant network). However, due to the limitation of network resources, a large number of message replicas generated by this forwarding mode will cause huge resource consumption and eventually lead to network congestion. A congestion control algorithm PAD. based on probabilistic admission and discard (probabilistic acceptance and drop,PAD is proposed. The algorithm combines queue length and input / output rate to detect congestion. Each node determines the probability of receiving and dropping messages according to the current congestion state, thus achieving lower overhead and higher message arrival rate. In addition, the continuous time Markov chain of message replicas is constructed based on the birth and death model, and the message arrival rate is analyzed theoretically. The theoretical analysis and simulation results show that compared with other algorithms, the PAD algorithm can significantly increase the message arrival rate by more than 130% while ensuring lower network overhead and shorter end-to-end delay.
【作者单位】：中南大学信息科学与工程学院;
【基金】：国家自然科学基金(61103204,61163060,61173169)资助课题
【分类号】：TN915.0

【参考文献】