Internet Service Time Distributions

Abstract

This thesis deals with the mathematics behind the speed of internet services and has both a theoretical and an experimental component. The goal is to give a better understanding of service time distributions and provide methods to predict its properties.

The theoretical part is written from an optimization perspective. It describes situations where a choice must be made from several service providers based on information about their service time distributions. This is done both for jobs that require a single service treatment and ones that require several services to work on them consecutively. It deals with problems such as incomplete information and service deterioration.

Subsequently we performed two series of experiments. The first consisted of measuring the response times for requests to actual internet services and analyzing the results. A comparison was made between several of the service time distributions and the effect of background traffic rate was shown by separating service times during the day from those at night. The second series of experiments was done in a lab environment where facets like background traffic could be regulated. Distributions of response times were analyzed, and those with different background traffic rates compared.