Traffic Driven Analysis of Cellular and WiFi Networks

Thumbnail Image
Paul, Utpal Kumar
Issue Date
Research Projects
Organizational Units
Journal Issue
Alternative Title
Since the days Internet traffic proliferated, measurement, monitoring and analysis of network traffic have been critical to not only the basic understanding of large networks, but also to seek improvements in resource management, traffic engineering and security. At the current times traffic in wireless local and wide area networks are facing similar upsurge. This calls for a similar vigor in traffic analysis studies. This thesis focuses on several traffic analysis studies in both cellular data networks and WiFi LANs. The broad goal is (i) to improve the understanding of the traffic dynamics, to explore structures in the traffic to help cost-effective monitoring and building of new traffic management strategies --- in the context of cellular networks, and (ii) understanding the interference properties and detecting misbehavior --- in the context of WiFi networks. We first use a large-scale data set collected inside a nationwide 3G cellular data network and conduct a detailed measurement analysis of network resource usage and subscriber behavior. We characterize subscriber mobility and temporal activity patterns and identify their relation to traffic volume. We also investigate how efficiently radio resources are used by different subscribers as well as by different applications. Our analysis using different statistical techniques shows existence of significant spatial correlation in radio resource usage in the base stations. We also use the concept of Granger Causality to understand the underlying functional connectivity and flow of influence in the network. Broadly, our observations deliver important insights into network-wide resource usage. Next, we propose a new traffic management technique for cellular data networks to improve networks' resource crisis situation in the face of exponential increase in mobile data traffic volume. Here we consider the existence of a higher-layer, agent-based scheduling system that could potentially delay scheduling of low priority flows at peak loads. The priorities are assumed to be user or application tagged, either automatically or manually. The general goal is to potentially move the low priority flows in time and space opportunistically to reduce the overall resource needs. We develop and evaluate two scheduling schemes. Simulation results using our large-scale cellular network trace data show the potential of these approaches in reducing base station resource requirements. Next, we present a scalable traffic measurement and monitoring technique for cellular data networks. We use a machine learning technique to learn the underlying conditional dependence structure in the base station traffic loads to show how such probabilistic models can be used to reduce the traffic monitoring efforts. The broad goal is to exploit the model to develop a spatial sampling technique that estimates the loads on all the base stations based on actual measurements only on a small subset of base stations. To understand the tradeoff between the accuracy and monitoring complexity better, we also study the use of this modeling approach on real applications. Two applications are studied --- energy saving and opportunistic scheduling. They show that load estimation via such modeling is quite effective in reducing the monitoring burden. In the last part of our thesis, we turn our attention to WiFi networks. We present a tool to estimate the interference between nodes and links in a live WiFi network by passive monitoring of wireless traffic. Our approach requires deploying multiple sniffers across the network to capture wireless traffic traces. These traces are then analyzed using a machine learning approach to infer the carrier-sense relationship between network nodes. We also demonstrate an important application of this tool--detection of selfish carrier-sense behavior. This is based on identifying any asymmetry in carrier-sense behavior between node pairs and finding multiple witnesses to raise confidence. We evaluate the tool using extensive experiments and simulation which demonstrate the effectiveness of both the applications.
168 pg.
The Graduate School, Stony Brook University: Stony Brook, NY.
PubMed ID