|
Abstract : |
Network tomography is a promising new technique for studying the (internal) behavior of large-scale networks based solely on end-to-end measurements. Most network loss tomography methods utilize active probing. While such measurement schemes are ecient, the probing burden may become prohibitive for large-scale networks. As an alternative, we propose a new, completely passive measurement framework based on sampling of existing TCP trac ows. The new passive unicast network tomography methodology we propose shows considerable promise. We demonstrate its performance using extensive ns-2 simulations. We observe that we are able to accurately estimate the losses experienced by existing TCP ows. As these can dier substantially from losses suered by other forms of trac, we surmise that in some situations inference from active probing may oer a poor re ection of existing TCP loss rates. We conduct theoretical queuing analysis which corroborates the results obtained through simulation experiment., |
