基于链路内在相关性的IP网络拥塞链路丢包率推断算法

IP Network Congestion Link Packet Loss Rate Inference Algorithm Based on Link Intrinsic Correlation

为解决网络链路丢包率推理算法中网络拓扑复杂、链路丢包率分析不准确等问题,已有研究采用假设子链路或通过率较高的路径中的链路作为不丢包链路,或者假设共享数目最多的链路为丢包链路,但是这种假设缺少有效的推理和证明。为解决此问题,提出了基于链路内在相关性的IP网络拥塞链路丢包率推断算法。该算法首先基于链路内联关系将网络模型化简并划分为多个独立子集;其次,对每个独立子集建立基于贝叶斯网络的链路拥塞推理模型,并基于每条链路的拥塞贡献率推理链路拥塞概率排序集合;最后,对每个独立子集,基于代数模型推理求解化简后的非奇异矩阵的唯一解,从而得到所有拥塞链路的丢包率。通过与算法LABLA和算法NTSPA比较可知,该算法具有较好的拥塞链路推理效果。

In order to solve the problem of complex network topology and inaccurate analysis of link loss rate in the network link loss rate inference algorithm,previous studies used a hypothesis that sub-link or a link in a path with a high pass rate is a non-loss link,or the most shared link is a packet loss link,but these assumptions lack valid reasoning and proof.In order to solve this problem,this paper proposed IP network congestion link packet loss rate inference algorithm based on link intrinsic correlation.The algorithm first decomposes the network model into multiple independent subsets based on the link inline relationship.Then,a Bayesian network based link congestion inference model is established for each independent subset,and the link congestion probability ranking set is inferred based on the congestion contribution rate of each link.Finally,for each independent subset,the unique solution of the reduced nonsingular matrix is solved based on algebraic model inference,and the packet loss rate of all the congested links is obtained.Compared with the algorithms LABLA and NTSPA,the proposed algorithm achieved better congestion link inference performance.