Network Link Loss Tomography Study

Network tomography can estimate the internal state of the network parameters by methods of statistical inference based on the collection of end-to-end related information, without the collaboration of the internal nodes. It can be used to estimate the link loss rate which can be exploited in routing, network optimization, fault detection, traffic engineering, network management and control and so on. In this paper, we apply the link-loss inference to fault detection and wireless sensor networks routing and propose two algorithms, one is neural network based congestion link diagnosis and the other is estimated link loss rate based directed diffusion routing protocol. The two methods have been validated by different simulation tools.Aiming to enhance the validity of wireless sensor networks, a new protocol which is named estimated link loss rate based directed diffusion routing protocol has been proposed. The main tasks include: (1)Since every node has the data fusion ability, when sensing data flow to sink node by using directed diffusion routing protocol, the logical topology of the network form a reverse multicast tree whose root is sink node. Through analyzing the received data by factor graph, the link loss rate can be estimated. (2)A sensing task is described using attribute-value name scheme, such a description is called an interest. During a fixed period, the link loss rate is inferred at sink node through the end-to-end data transmission information, then the interest message which contains the link loss rate will be broadcast to the entire network periodically. (3) In the stage of gradient setting up, the node which receives the interest will establish the gradient to the node which sends the message and records the loss rate information of each neighbor node from the interest message. (4)The data packets are transmitted along the link with the lowest link packet loss rate and delay. (5)This algorithm can improve the network quality of service in aspects of packet loss rate and link delay, and the simulation results show that this protocol is superior to directed diffusion protocol in terms of end-to-end average packet loss rate.Link loss rates tomography can also be applied to finding the congestion link. In this paper, an algorithm which called neural network based congestion link diagnosis has been proposed. The main tasks contain: (1)Through in-depth studying of the SCFS(smallest consistent failure set) algorithm and the CLINK(congested link) algorithm which are two classic congestion link diagnosis methods, we find that the prior probability is the precondition of inferring congested links set. (2) The prior information of link status is obtained by training the neural network with the historical data. (3) Exploiting the trained network, input a new end-to-end measurements to diagnose the true congested link set. (4)Compared with the CLINK algorithm, the simulation has shown that the algorithm we proposed has a good performance in finding the congestion link.