Project

data streaming; peer-to-peer computing; distributed algorithms; reliability; scalability; overlay networks; peer-to-peer; distributed systems; multicast; large-scale systems

Lead
Lay summary
Recent advances in computer networking and high-speed data transmission technologies have led to considerable increase in the average bandwidth available to end-users. As a result, since the introduction of peer-to-peer file sharing systems, a number of information diffusion applications have emerged, notably in the context of media streaming (i.e., audio and video streaming). These applications typically rely on some support for data multicast, where peers interested in a given data stream can join a corresponding multicast group. Although application-level multicast (also known as overlay multicast) is not as efficient as native IP multicast, it is easier to deploy and manage, and therefore has become very popular in multimedia streaming for largely distributed services and applications. Overlay networks are typically composed of one or more propagation trees or a single mesh. In these structures, nodes are computers and edges are overlay links formed by the establishment of peering relationships between nodes. This project intends to investigate a separation-of-concerns approach to the media-streaming problem. The idea is to decompose data dissemination into two parts, one focusing on distributed techniques to build meshes efficiently and the other on sophisticated data propagation structures. The resulting architecture is quite modular, composed of two layers.Every year Internet enterprises and content providers spend hefty amounts of money on hardware, maintenance, power supply and cooling equipments in response to the ever-growing amount of available information and increasing number of information-thirsty users. In addition to such direct costs, there are substantial environmental impacts caused by this expanding operation. The primary goal of peer-to-peer systems is to eliminate the need for central servers and harnessing the underutilized bandwidth and processing power of machines on the network. In such a system, every ordinary device or computer with a mediocre Internet link can function as a tiny content provider contributing some storage and bandwidth. With a few thousand nodes, this small contribution can sum up to huge amounts of resources ready to serve large contents without the need for centralized content providers and their inherent costs.This research is an effort to fill the gap and find ways to make overlay multicast an efficient, inexpensive, and clean alternative to conventional distribution techniques.

Direct link to Lay Summary

Last update: 21.02.2013

Title	Year

Number	Title	Start	Funding scheme

Recent advances in computer networking and high-speed data transmission technologies have led to considerable increase in the average bandwidth available to end users. As a result, since the introduction of peer-to-peer file sharing systems, a number of information diffusion applications have emerged, notably in the context of media streaming (i.e., audio and video streaming). These applications typically rely on some support for data multicast, where peers interested in a given data stream can join a corresponding multicast group. Although application-level multicast (also known as overlay multicast) is not as efficient as native IP multicast, it is easier to deploy and manage, and therefore has become very popular in multimedia streaming for largely distributed services and applications. Reliance on ordinary peers as active participants in an overlay network, however, brings its own challenges.Overlay networks are typically composed of one or more propagation trees or a single mesh. In these structures, nodes are computers and edges are overlay links formed by the establishment of peering relationships between nodes. This project intends to investigate a separation-of-concerns approach to the media streaming problem. The idea is to decompose data dissemination into two parts, one focusing on distributed techniques to build meshes efficiently and the other on sophisticated data propagation structures. The resulting architecture is quite modular, composed of two layers. The purpose of the first layer is to maintain a connected structure and to provide a basic building block on top of which different data-streaming protocols can be deployed. Different mesh construction algorithms can be devised to implement this layer. In the second layer, nodes build the data propagation structure (e.g., a tree). To do so, nodes gossip topology updates, which eventually gives them an approximate view of the network, either global or partial. When the data source is ready to stream data, it first computes the propagation structure and sends it to its children, which will recursively do the same until all nodes are aware of the structure. Some preliminary research we have conducted in the scope of this project has shown that while our approach seems promising, it raises a number of challenges, as we illustrate in this proposal.Using graphs to explain behavioral properties of social, biological, and technological networks has been considerably popular recently. Social scientists and biologists, for example, usually study existing networks shaped by human relationships and natural phenomena. In contrast, we aim at identifying desired and undesired graph properties and at devising algorithms to produce networks exhibiting the former properties and avoiding the latter. To achieve this, two challenges come into play. First, the underlaying physical topology and behavior of the end users impose limitations and restrict design choices. Second, for scalability, the solution must be distributed and decentralized with a limited knowledge about the network. While classical graph theory has been around for centuries, more recent trends in the analysis of graphs and their properties have led to the emergence of a number of techniques that could contribute to this project, in particular along the line of modeling the dynamic aspects of an evolving topology.Finally, to adequately impact the state of the art in its field, we believe that the project must look into both theoretical and systems aspects of the problem, that is, we intend to conduct our research in sound theoretical ground, and implement and experimentally evaluate our architecture.