• Lars Møller Mikkelsen
  • Michael Sølvkær Pedersen
  • Steffen Riber Thomsen
4. term, Networks and Distributed Systems, Master (Master Programme)
This project is a continuation of the work on the NetMap system, which is a network measurement system developed for Android devices, aimed at using crowdsourcing for gathering measurements, to be used in creating a Network Performance Map.
In this project there are 2 main focus areas: Extended analysis of the measurements gathered with the NetMap system, and optimization of measurement methods.
The extended analysis includes analyzing the impact of the following points on the connection performance: movement speed, handovers between cell antennas, area types, and device models.
Furthermore, the currently implemented methods for measuring Round Trip Times (RTTs) and throughput are analyzed in-depth to get a deeper understanding of the communication taking place during the measurement.
In the optimization of measurement methods, 2 techniques for estimating the available bandwidth, while keeping the data usage at a minimum, are implemented in the system and tested on a real cellular connection. Results from these techniques are compared to results from the current throughput estimation method, and evaluated based on these circumstances.
From the extended analysis it is concluded that when a device moves faster a slight decrease in performance is seen, and that in areas where handovers tend to happen the performance is decreased as well.
Furthermore, it is shown that the performance typically is better in small town areas than in urban areas, believed to be due to the combination of fewer users and more network resources.
From the analysis of device models, examples of differences in performance between high and low-end devices are presented.
Based on the analysis of the current measurement methods, some alternative methods for extracting more detailed results are suggested.
Two additional bandwidth estimation techniques have been implemented and analyzed. It is concluded that one technique delivers satisfactory estimations while keeping the data usage extremely low, where the other technique experienced issues due to the requirement of time precision in transmission of measurement packets.
Publication date6 Jun 2013
Number of pages164
ID: 77274189