| The network tomography is concentrate on the end-to-end measurement to inferthe network link parameters. The network tomography is an efficient method when thenetwork internal switches or routers are unable to measure statistical information ofnetwork interior links directly. Although network parameters inference technology ofmulticast has lower bandwidth and strong relativity of packets etc, the presentmajority networks do not support multicast,therefore, the unicast tomography becomesone hot spot in this area.Although collaboration and cooperation is not needed in network tomographybased on unicast, but strictly one mapping relations can’t be guarantee in the path andlink of unicast as multicast, which brings a range of parameter estimating problems ofnetwork link, such as: the emergence of non-routing matrix, the inefficient contractingprobes, the high-dimensional Diophantine equation and so on. In order to solve theabove problems, we propose unicast-based tomography to infer the network link delaydistributions and loss, the main work is as follows of two aspects:Firstly, the cumulant generating function is adjusted to correct the estimationerror according to the link delay distributions for the estimation error. In this method,the link delay of cumulant generating function(CGF) is used to infer the link delaydistribution estimates and error correction. As is showed in simulation result, asmaller mean square error is achieved.Secondly, according to the problem of low efficiency in the current probes, wemake use a way of tomography based on unicast packets. By sending probing packetsfrom the root node to all leaf nodes, the adjacent node or nodes in an ensemble ofsimilar groups into packet pair, three-packet stripe, four-packet stripe, whichconstitute overdetermined equations, loss and condition of loss can be solved byequations. In this way, measurement accuracy can be guaranteed under the premise ofimproving the accuracy of solution, stability and efficiency of contracting award. |
PDF Full Text Request