Within our ongoing research work, we frequently have open sub-problems that can be assigned to VU students as masters theses. Examples of completed master theses within our group can be viewed here.

Master theses will all have a research aspect and a practical aspect. Candidates should therefore have good network and distributed systems background as well as reasonable C/C++ programming skills.

Are you interested in joining us? Have a look at the following list of subjects and contact Guillaume Pierre. (Bear in mind that subjects listed in this page are often outdated. Take them as examples of the type of subjects we usually offer, and contact us for more up-to-date information.)

The Globule project is working on a complete redesign of the hosting platform of the Wikipedia website. The goal is to host Wikipedia on a fully decentralized system. About a year ago we wrote a paper describing this peer-to-peer architecture. However, at the time we wrote that paper we explained why we thought the design made sense, but we did not have any hard data to validate these claims. So, we contacted Wikipedia and finally obtained an access trace of the actual Wikipedia website from them (note that this trace is huge: dozens of billions of entries!). More information that trace can be found here.

I am looking for a student who could conduct experiments to validate (or invalidate!) our decentralized architecture based on real data. The project would require to build a simulation that is scalable enough to simulate the entire decentralized wikipedia, then to study how the current design would work under the workload we got from the traces, and if necessary propose changes to the decentralized wikipedia system design to better accomodate that traffic.

BitTorrent allows peers to exchange file content with each other. However, the selection of peers with whom one may interact is mostly random. This means that any client may interact with any other. However, this has two negative implications: (i) the communication between an arbitrary pair of peers may be pretty bad; (ii) ISPs complain because long-distance traffic is more expensive for them than short-distance traffic. The goal of this project is to investigate how we can create a bias in the selection of peers towards closeby nodes, and what the potential benefits may be. One potential outcome of this project could be an extension to the Azureus bittorrent client, that one may name an "ISP-friendly Azureus".

I may have more topics available that for various reasons did not make it onto this Webpage. Feel free to ask if you would like to know more!