A model for the marginal distribution of aggregate per second HTTP request rate

This paper presents a new model for aggregate hypertext transfer protocol (HTTP) request rate. Specifically the model describes the marginal distribution of aggregate HTTP request rate at the second time scale for the aggregate Web traffic generated by a large number of users accessing the Web. The examination of three independent traces of Web traffic shows that the marginal distribution of HTTP request rate is well modelled by the Polya-Aeppli probability distribution. The Polya-Aeppli result is based on observations that the marginal distribution of active users is Poisson and that the distribution of the number of requests generated by an active user is approximately geometric. The Polya-Aeppli result has immediate application in the estimation of peak HTTP request rates from known mean and variance. The result also highlights a discrepancy between artificial Web traffic workloads used for cache benchmarking based on the Poisson assumption and actual Web traffic.