Home > Events > PhD Defense - Lixing Song

PhD Defense - Lixing Song

Start: 6/25/2018 at 2:30PM
End: 6/25/2018 at 5:30PM
Location: 258 Fitzpatrick
Attendees: Faculty and students are welcome to attend the presentation portion of the defense. Light refreshments will be served.
Add to calendar:
iCal vCal

Lixing Song

PhD Dissertation Defense

        June 25, 2018        2:30 pm        258 Fitzpatrick

Adviser:  Dr. Aaron Striegel

Committee Members:

 Dr. Richard Billo         Dr. Christian Poellabauer        Dr. Dong Wang




The Internet today is actively embracing the evolution of mobilization. In 2017, mobile devices consumed 68% of overall Internet traffic. This number is expected to grow sevenfold in the next five years. Due to the limited resources on wireless spectrum, the relentlessly growing mobile data demand can portend an ominous future for the Quality of Experience (QoE) on mobile networks (e.g., WiFi and cellular). In addition, with the dynamics of the wireless channels the performance on mobile networks tends to vary significantly often making the wireless link​ as the bottleneck to decide QoE. Network characterization is a tool to understand the performance of mobile networks. Unfortunately, existing solutions are either cumbersome or inaccurate. Some are effective but the cost is prohibitively expensive (e.g., cost tens of megabyte data and take the order of tens of seconds to finish). Other methods are lightweight but yield low accuracy. Critically, the different transmission schemes under modern mobile networks make conventional characterization methods fail due to the issues with the lower-layer behaviors. Therefore, the key challenge is how to conduct mobile network characterization in an accurate and efficient manner.  

In this dissertation, I propose a test suite of mobile network characterization using both active and passive approaches on WiFi and cellular networks. By carefully studying the data transmission behaviors at the lower layers (e.g., the physical layer), my approach manipulates the observations of traffic patterns on the upper layers (e.g., the transport layer) to enable accurate and efficient characterization methods. Specifically, in this dissertation, I design two active available bandwidth estimation tools focusing on WiFi and LTE networks.  By leveraging the aggregation/batching properties on WiFi and LTE, I design a probe packet train that utilizes the concept of self-induced congestion. I implement the estimation tools in an HTTP-based platform---FMNC (Fast Mobile Network Characterization). This system is designed to transmit packet sequence in a sliced, structured, and reordered manner. This feature enables mobile users to run the designed tests without installing a dedicated application on the client side.  In addition to the active approaches, I also propose an efficient passive client-side traffic characterization method on WiFi networks. By exploiting the frame aggregation feature on WiFi, the method can piggyback on a WiFi scan operation and achieve accurate traffic characterization with minimal traffic capture (i.e., control traffic only). I conduct two case studies using the passive characterization method: a real-world measurement case and an application case in video streaming. The measurement study reveals interesting observations under different network scenarios. The application case helps improve stall rate of video streaming under WiFi. Overall,  all the proposed methods/applications in this dissertation have been carefully evaluated through extensive in-lab and real-world experiments.