Web traffic modeling exploiting TCP connections' temporal clustering through HTML-REDUCE

This article presents a measurement analysis methodology that, starting from packet-level traces, identifies some statistics useful to characterize Web traffic, such as page size, page request frequency, and the user's think time between the download of pages. The methodology was implemented in a tool named HTML-REDUCE and applied on traces collected in CSELT's corporate network. For each identified statistic the analytic expression best approximating its empirical distribution is searched. A simplified model for bandwidth dimensioning in the server-to-client direction is then presented and validated through comparison with full protocol stack simulations. The model only requires the page request frequency of an aggregate of clients and the first two moments of page size distribution.