Research

The goal of our research is to create, understand, measure, and improve distributed online economies. Tribler is our multi-year case study focused on Internet TV. This is motivated by our belief that the best method to improve Internet TV is creating an on-line economy around the involved computer resources. With our research we hope to make a contribution to the predicted big upcoming changes in the TV landscape. We use an open source software development model and everyone is welcome to contribute to our non-profit initiative by downloading the source code.

Currently, we are tackling the following research challenges with the goal to eventually implement them into Tribler:

HDTV Video-on-Demand
In an ideal P2P system, users obtain a real-time data stream using the full download capacity of their Internet connection. Video quality varies automatically depending on the quality of the Internet connection. The upload capacity no longer forms a bottleneck.
Wealth of content
In an efficient system video creators and consumers are tied into the same on-line economy. Direct dealings create efficiency with access to both rare long-tail content and user-generated content.
Rich metadata
With sufficient trust and robustness users can enhance the metadata of content. Contributions range from simple subjective quality ratings to more advanced tagging, subtitling, and comments. Instead of adding tags to a complete video, support is added to index, tag, or comment on a specific scene in a longer on-line video.

Our research uses pure P2P architectures, meaning that there is no central entity for any functionality. This is unlike previous BitTorrent implementations that rely on centralized trackers. Our fully decentralized approach guarantees resistance to flash crowds and provides unbounded scalability.

Bandwith as a global currency

Our first step towards creating an online economy is creating a marketplace around bandwidth. Tribler V4.1 is a first attempt towards turning bandwidth into a global currency. Sophisticated mechanisms such as dynamic pricing of resources to match supply with demand are still poorly understood and cannot yet be exploited in Internet TV.

Implementation: BarterCast and Give-2-Get

Tribler V4.1 goes beyond the BitTorrent tit-for-tat protocol. The tit-for-tat algorithm can best be described as a 'temporally local method without long term credits'. In the current Tribler version, the locality limit is removed, scaling beyond two peers in the same swarm is allowed and long term credit accumulation is enabled. We implemented a distributed accounting framework for bandwidth which is fully backward compatible with the BitTorrent protocol. For future Tribler versions we aim to realize secure accounting to make bandwidth a global currency.

Our solution uses a two-step approach at the micro and macro level with one underlying security measure that prevents identity theft. We added non-spoofable identities to the Bittorrent protocol using Elliptic Curve Cryptography (Python code) and a new challenge/response handshake. On top of the non-spoofable identities we added a protocol to exchange the 'altruism levels' of peers, called BarterCast (Python code). Here, altruism is defined as the amount of uploading versus downloading of a user. The protocol operates at the macro level addressing the problem in BitTorrent that there is no incentive to seed a file once a download is complete. For further details please see this draft of a full technical description.

At the micro or swarm level we implemented the Give-to-Get algorithm (Python code). This algorithm is designed for Video-on-Demand, which cannot be supported with tit-for-tat in combination with the default piece picking policy. Give-to-Get provides more upload bandwidth to the peers which also donate upload bandwidth. The underlying principle of tit-for-tat is that bandwidth is bartered with the same peer. Give-to-get on the other hand also supports forwarding: receiving a chunk from one peer and passing it on to the next. By communicating at the micro-level about provided and obtained bandwidth and favoring helpful peers we ensure that leeching is not productive. Video-on-Demand service becomes available when sufficient peers in a BitTorrent swarm support this protocol extension. More details and performance evaluations can be found in our paper 'Give-to-Get: an algorithm for P2P Video on Demand (PDF)'. This work will be presented at the Fifteenth Annual Multimedia Computing and Networking conference to be held in January 2008 in San Jose, California.

As a result of BarterCast and Give-to-Get each peer can calculate which peer has proven to be helpful in the past. In combination with the integrated social network in Tribler (friends, friends-of-friends) we aim to create a fully distributed trust system in the next Tribler version. Such a trust system would enable ratings, tag-based navigation, moderation, and rich metadata. The underlying principle of using your friends and active participants in the system creates a powerful 'web-of-trust' to prevent many type of attacks. Creating a robust and fully distributed trust system is a very desirable research goal in particular given current levels of spam and abuse on the Internet.

BitTorrent protocol exploits

Our research on distributed trust systems focuses on preventing active attacks, fraud, and abuse of the system. A distributed on-line economy needs to have a tight security to ensure a functioning marketplace and prevent stealing of resources or currency.

Bitthief and Bittyrant are two research projects that have demonstrated exploits for the BitTorrent protocol. We present a security framework to counter such exploits. With BarterCast and Give-to-Get in place, Tribler is resistant against the attacks demonstrated by Bitthief and Bittyrant.