Wednesday, 29 June 2011

Master thesis - Abstract

I am going to present my master thesis on 30th of June, at 8:30 of the morning. The formal information are presented below. Anyone is welcome to see the presentation and ask questions.

Student: Jiahao Liu
Title: Monitoring of user experience and interactions for mobile video streaming services
Time: Thursday 30th of June at 08:30
Place: Seminar room Palo Alto, Isafjordsgatan 22 (Elevator C, 4 floor), Kista
Examiner: Professor Zary Segall
Supervisor: Pietro Lungaro
Opponent: Julien Michel Guy Dubois
Language: English


Mobile data traffic is expected to continue growing at an almost exponential rate in the next five to ten years. Contributing to this trend are video streaming services, e.g. YouTube and Netflix, which are currently consuming significant portions of the available bandwidth and are likely to remain very popular services in the near future.

In order to meet future traffic demands, mobile operators can either continue with the current approach consisting of proportionally increasing the infrastructure deployment, or adopting some alternative and opportunistic paradigms for content delivery, capable of exploiting the temporal and geographical traffic variations while optimizing the amount of resources invested in content delivery.
While the former approach is likely to have only very limited economic feasibility, due to substantial increases in OPEX and CAPEX, the latter is instead very promising but require experimental validation and new software tools for performing more informed resource allocation decisions. A series of initial experimental investigations have been performed within the COSEM project, performed at Wireless@KTH, to assess the feasibility of context-aware content pre-fetching in mobile networks. The initial results showed that an unprecedented increase in user experience, together with terminal energy cost reduction and increased network resource utilization, can be achieved with pre-fetching, provided that future content requests can be predicted with some degrees of reliability.

To understand to what extent content can be predicted and how to utilize user context information for increasing the reliability of the predictions, this master thesis proposes an innovative monitoring approach, in which a software application, running "on the background" of mobile devices, is designed for collecting information on both the user context and her interactions with the mobile device and the various running services. In particular, the main focus of this Thesis has been on capturing user interactions with the YouTube mobile application and monitoring various context information associated with video content consumption. The monitored information includes a set of service-related variables such as location and time of access, buffering and QoS-related information, video specific information such as video tags, ID and categories and also a series of QoE and user-interaction variables, including the occurrences of video interruptions and user actions within the various “states” of the YouTube app.

The results show, with one of the methods chosen, for the YouTube application running in Android System and without any direct interaction, 90.51% of the transition between states and 95% of the internal states can be monitored. The 9.49% of the transition between states that cannot be monitored corresponds to showing the keyboard on the screen, which has minimal importance; the 5% of the non-monitored internal states corresponds to the change on the settings of the application, which can be monitored with other methods.
Furthermore, in order to improve the univesality of the information obtained, making it easily to be shared and accessed, all the data are saved in RDF format, which contains an uniform resource identifier (URI) making each resource unique.

No comments:

Post a Comment