Bottleneck location estimation for scalable multicast using maximum likelihood estimation

Abstract

The efficiency of end-to-end multicast transport depends critically upon their ability to scale efficiently to a large number of receiver. Also, the QoS requirement for various multimedia service is so stringent that it makes the formation of the distribution tree difficult. Designing a reliable multicast-based network that scales to the size of multicast group member is a difficult issue because of the diversity of user demands to meet. Therefore we wish to build a foundation for QoS-guaranteed multicast tree and apply the spectral analysis which characterizes the uniqueness of the source traffic. However, to make multicast service reliable accurate measurement schemes should precede. While there are many development projects on unicast-based measurement, several multicast-based end-to-end measurement schemes is being watched with keen interest. The inference of multicast tree and loss inference are included. The inference of multicast tree is executed by the shared loss patterns and needs no explicit knowledge of the tree. This approach relies upon complete information of loss statistics at every receiver and thus is not a practical protocol building block in its own right. On the other hand, loss inference of internal nodes because they use the MLE(Maximum Likelihood Estimator). But they know the multicast distribution tree in advance. Schemes using MLE are very efficient and the inferred value converges fast and surely to its true value. In the theoretical analysis, internal delay estimation is possible but their analysis is so complex due to the continuity property of the delay. In the thesis, we propose the BLE(Bottleneck Location Estimator) which can overcome the analytic difficulty of the delay estimation using the power spectrum of the packet interarrival time as the performance metric. We develop the basic model of inferring bottleneck location and show the efficiency of the BLE by both of the theoretical analysis and simulation. In terms of sca...