Lars Møller Mikkelsen

NetMap - Creating a Map of Application Layer QoS Metrics of Mobile Networks Using Crowd Sourcing

Based on the continuous increase in network traffic on mobile networks, the large increase in smart devices, and the ever ongoing development of Internet enabled services, we argue for the need of a network performance map. In this paper NetMap is presented, which is a measurement system based on crowd sourcing, that utilizes end user smart devices in automatically measuring and gathering network performance metrics on mobile networks. Metrics measured include throughput, round trip times, connectivity, and signal strength, and are accompanied by a wide range of context information about the device state. The potential of the NetMap approach is demonstrated by three usage examples.

Performance Evaluation of Methods for Estimating Achievable Throughput on Cellular Connections

The continuous increase in always connected devices and the advance in capabilities of networks and services offered via the network is evident. The target group of most of these devices and services is users, so how users perceive the network performance is of great importance. Estimating achievable throughput (AT) is the main focus of this paper, which can be expressed as the data rate that users experience. We establish the Bulk Transfer Capacity (BTC) method as the ground truth of the AT. We choose to evaluate the Trains of Packet-Pair (TOPP) method as an alternative to BTC in estimating AT, due to its much reduced resource consumption. Based on real-life measurements of the two methods we conclude that TOPP is a good candidate to estimate AT, based on similarity in results with BTC.

Enhancing IoT Systems by Exploiting Opportunistically Collected Information from Communication Networks

In this thesis, we investigate how to enhance internet of things (IoT) systems by utilizing information about network performance and about number of people in a limited area. Network performance information is relevant for distributed IoT systems as they are affected by the performance of the network between their components. Knowing about the performance of the network allows these systems to optimize scheduling of communication. People count information is relevant for IoT systems because they make out the link between the digital and physical world interacting with users. Knowing about the number of users in an area translates to into the load or availability of the system. With these information types available, IoT systems can be made truly smart and adaptable to the context they operate in, and not just operate based on a predefined set of rules or threshold values. The main challenge in obtaining these information types is how to measure and collect them. This is because they are highly dynamic as they are influenced by the current situation at the location where they are measured. In this work, we propose how to obtain these parameters cost effectively by opportunistically collecting them using automated approaches. Specifically, network performance information is collected using crowd sourcing measurements and people count information is obtained as estimations based on passively collected WiFi probes. Furthermore, we propose how to exploit the collected information types in a real IoT system. The thesis consists of three main contributions. In the first contribution, we present NetMap, a system we developed for performing and collecting transport layer end-to-end network performance measurements. NetMap exploits the ubiquitous presence of smartphones to collect many measurements at low cost, which are used for populating a network performance map with real network measurements. We focus on the network performance between end user device and back end system, including the 3G and 4G cellular connections and their influence on the application traffic. Specifically, we look at the metrics round-trip time and achievable throughput in estimating end-toend transport layer network performance, but we also include received power measurements. In doing this we perform an evaluation of two methods for

Location Assisted Vertical Handover Algorithm for QoS Optimization in End-to-end Connections

When devices are mobile, they potentially move within range of several different wireless networks, which are not utilized due to security, inactive radios or it is just plain cumbersome for the user to exploit them. This paper proposes a vertical handover algorithm to utilize these networks, which selects the best available network at a given time, factoring in prediction of user movement, energy consumption, QoS of connections and application demand for the end-to-end connection in a (multiple) client/server setting. The algorithm allows support for different mechanisms of interaction, where we in this paper focus on 1) peer-to-peer in a WLAN setting, 2) p2p behind NAT and 3) what we call a server bounce mechanism. The algorithm is supported by a User-specific Virtual Network to obtain required network state information. Experimental tests are conducted, using both simulations and actual implementation on Android based tablets. The simulations cover a wide range of scenarios for two mobile users in an urban area with ubiquitous cellular coverage, and shows our algorithm leads to increased throughput, with fewer handovers, when considering the end-to-end connection than to other handover schemes. The implementation utilizing real world networks, shows the feasibility of the algorithm due to its low complexity compared to other algorithms.