Study On Node State Independence In Autonomous Systems

Providing QoS guarantee in packet-switching network has been a key issue these years. It is of great importance and difficulty to provide end-to-end QoS guarantee in Access Networks. Most of the present QoS mechanisms in Access Networks are based on IntServ model. It is connection-oriented and its per-flow resource reservation and states keeping scheme can't accommodate the future development of network application. The author illustrates the advantages and preconditions for the establishment of connectionless statistical QoS guarantee. Requirements to choose delay Probability Density Function (PDF) as state information is explained and node state independence assumption is chosen as the subject of this thesis.Link-delay independence assumption is frequently used in QoS research work because it can make many questions less complex. This assumption is extremely important when link-delay PDF is used as state information because the calculation of path-delay PDF is NP-Hard without it. If the node state or link-delay is independent and their delay PDF is available, then all the path-delay PDF in the network can be calculated by the convolution of link-delay PDF and each path metric, such as average delay, delay variation and loss probability, can be achieved easily. According to our investigation, there has no verification for this assumption in Autonomous Systems.A Source Routing, also called State Routing, simulation platform with MCI topology is set up to study the independence of node states in Autonomous Systems with self-similar traffic. A quantized probability distribution PF is used instead of PDF. In our simulation, link-delay PF is obtained from the periodical sampling of queue length of each link. Path-delay PF is obtained from probe packets periodically sent through the tested path. This is a kind of system-wide observation and all the measurements are independent with the traffic. In order to give a comprehensive study of node state independence assumption, all of the link-delay PFs and 15 path-delay PFs in the network are tested. The same 15 path-delay PFs are also obtained by the convolution of above tested link-delay PFs. We measured the statistical distance between the above two sets of path-delay PFs by Kullback-Leibler divergence and further demonstrate its influence on the effectiveness of path metric estimation by the calculation of delay expectation, variation and loss probability. Our Simulation and analysis results indicate that node states are nearly independent in Autonomous Systems. Path metrics calculated from this assumption approximates to original values and can be used in multi-constrained QoS routing when link-delay PF is used as node state.