Extreme value engineering for local network traffic

Small switching offices and office information systems share some common properties in the description of load peaks. In both cases, because of the smaller user population, the structure of the underlying random process becomes important, and while the methods of extreme value engineering are applicable, the usual asymptotic distributions are not. The family of gamma distributions is proposed as a flexible process model (which does not imply negative load values), and it is used to model data in the literature on the traffic in small switching offices. New data from trials of office information systems yield extreme values of loads which support this proposal. The gamma model supports an extreme value engineering procedure for the design of peak monitoring methods, the calculation of return period loads, and the calculation of probability curves for extremes.<
>