FYP 2009-2010 (C.L. Wang)

Location information is essential to enable location-based services (LBS) in a metropolitan environment (e.g. Hong Kong). Innovative LBS can makes people’s life more easier and enjoyable. A typical service is Location-based advertising, which include:

(1) find nearby point of interest (POI) (e.g., restaurant, supermarket, clinic, shopping mall, cinema, hotel)

(2) sort advertisement's relevancy according to different factors (user preference, pricing, etc)

(3) then suggesting the shortest path to destination.

To localize a mobile device's location (especially in a city-like scale), common existing approaches are Wi-Fi localization (Skyhook), GPS, provider's localization (GSM tower ID). However, to provide a compressive yet useful localization solution, which is of reasonable accuracy (from 20 to 100 meters) at most of the time during a user's daily activity span, it requires:

Smoothing switching between different available localization modalities: e.g. switching from out door GPS to indoor Wi-Fi approach or switching from coarse-grained GSM tower/Wi-Fi information to fine-grained GPS location information

Opportunistic collaboration between mobile users that share partial location information, e.g. a non-GPS mobile phone can determine its location if it has at least 3 nearby GPS-enabled phones, whose GPS data are exchanged using bluetooth. (See the following figure)

Moreover if people can help each other calculate their own location with less support of third party providers (GSM tower localization), it will be a privacy plus that encourage people to adopt ad hoc localization in urban environments. This FYP project explores the feasibility of GPS-calibrated ad hoc localization in urban environment, which a small percentage of GPS-enabled mobile phones can help calibrate both the trajectory estimation and real-time localization of other mobile phones, by urban pedestrians in a metropolitan environment.

Implementation details:

(1) devise and implement a GPS information sharing/calculation algorithm in C++/C on the emulator

(2) put it in a mobile scenario simulation tool to record localization accuracy for certain durations

(3) give possible analysis of your method that explains localization accuracy

(4) port your algorithm as an application on Google's Android phone emulator/phone in Java

MobiREAL simulator : http://www.mobireal.net/
Google Android : http://code.google.com/android/

Project 2: BetterLife 2.0: Personalized Recommendation Service on Cloud

Supervisor: Dr. C.L. Wang
Number of students: 3-4

KONG Kwai Yee Ruby <kong.ruby@gmail.com>,

LO Fung Alvin <honye.lo@gmail.com>

WONG Kwok Kit Henry <henry1987@gmail.com>

Introduction:

With the popularity of social network websites that deal with certain type of user interest (e.g. music, friends, food, video, picture), much user generated data has been stored in these web 2.0 websites. One the other hand, more and more Internet connected mobile phones also generate data during their interactions with the Internet. For example, a customer can comment on a restaurant's meals using his mobile phone, which could also be uploaded to the web, and potentially used by other users.

This FYP project targets to make use of social networks to produce personalized public services (e.g. hotel, bus schedule, traffic, shop mall, restaurants) to mobile users, with real or synthetic environment data of Hong Kong. It treats data at the unit of user, which covers all aspects of his interests. This is in contrast with many social network sites that fractioned user data to its own interest (i.e. the book recommendation website only records a user's reading history). Since we treat all aspects of user data together, also potentially scale to many entities (e.g. 7 million people + 1 million public organizations), we need to explore reasoning mechanisms upon large-scale data using the powerful Cloud technologies.

Design & Implementation details:

Replace a web 2.0 website's database with the Cloud storage system, which has a large data size.
Apply reasoning techniques per user (e.g. pair-wise user data similarity comparison) to suggest certain aspects of interests (e.g.
popular music, nearby good dining spot, etc.)
Adopt its computation logic to Google's MapReduce paradigm. This requires design and analysis of computation breakdown, data partition/
dependency.

Tools:

(1) Case-based reasoning framework: jCollibri, may need to port its similarity matching algorithm
into parallelism and use the MapReduce computing mechanism the cloud computing infrastructure

(2) Cloud computing platforms:
(a) data storage: this can use subscribed Amazon's EC2/Sun Grid Engine or use Eucalyptus to set up your own cloud infrastructure.
(b) MapReduce programming: using Yahoo's Hadoop & HDFS

(3) Front-end user interface:

Web interface: use a social network website template (e.g. Elgg) for web front-end that allows user to login, upload cases, edit user data that will be stored into per user cloud storage (hadoop HDFS file system)
Mobile device interface: develop client application of Google Android Phone to upload/modify user data, and receive server side recommendation

Reference:

Social network: http://en.wikipedia.org/wiki/Social_networks

Elgg: an open source social network template http://elgg.org/>

MapReduce : http://code.google.com/edu/parallel/mapreduce-tutorial.html

Hadoop and Hadoop Distributed File System: http://en.wikipedia.org/wiki/Hadoop

Amazon Elastic Compute Cloud http://aws.amazon.com/ec2/

Eucalyptus: http://www.eucalyptus.com/
Andriod: http://code.google.com/android/

PS. . We will give tutorial on social network setup, Yahoo's Hadoop/MapReduce programming, cloud storage setup, jCollibri, and provide source code of similar Android client applications.

Project 4 (CE): Green Data Center

Supervisor: Dr. C.L. Wang
Number of students: 2

Data centers consume enormous energy nowadays (3M US Dollars in 2006 for Oak Ridge National Lab). Studies show that only less than 50% of the power is used for computation. However, cooling system accounts for another one-third of the overall power consumption. Therefore it's of vital importance to reduce the power consumption in modern data centers, both for the sake of cost reduction and environmental protection. Task migration is believed to be an effective way of redistributing tasks on overheated computers such that the heat-imbalance in the data center can be smoothed, which will result in substantial reduction of energy consumption on the cooling system. Students who take this project will need to:

Study how to monitor the heat distribution in a data center
Make use of Live Virtual Machine Migration technique to relocate running tasks from overheated computers
Analyze the effect/result of heat-redistribution and/or power reduction after adopting the implemented solution.

Benefits from this project: (1) knowledge and experience of large data center administration. (2) modern knowledge of task re-distribution technologies (e.g., Xen, VMware, VirtualBox). (3) learn Linux system/OS related skills

Reference:

Cullen Bash and George Forman, "Cool Job Allocation: Measuring the Power Savings of Placing Jobs at Cooling-Efficient Locations in the Data Center", Proceedings of the 2007 USENIX Annual Technical Conference, pp. 363–368.
Xen: http://xen.org/
Live Migration of Virtual Machines (pdf)