E D O S
Environment for the Distribution of Open Source Software
The project deals with the management and distribution of Open Source Software, developing theoretical and practical solutions to the management of large-scale, modular software projects, and their distribution.
The aim is to tackle:
Research and Academic Institutions
University of Paris 7, France
EDOS is divided into six work packages
Tel Aviv's part
Mainly WP4 - novel distribution of code over the internet.
Our main challenges
WP4 General Architecture
Peers are connected through an overlay which implements the following application structure:
Current people at TAU
Tal is currently studying the first layer of the structure; "Event notification". His main effort is to develop a new event notification system which utilizes a unique clustering feature, one that can be found in EDOS environment (and in other environments as well). The new application implements a novel algorithm designed in Tel Aviv University. Its framework is based on another event notification system called Scribe and on a DHT implementation called Pastry. The main algorithm aims to cluster different available topics in a dynamic and distributed fashion way, such that the total network maintenance overhead will cost as low as possible. The technique is based on the concept of aggregating related topics under one common topic (i.e. turn them into a group), and subscribing to its representative.
Itay is currently working on building the first edos prototype in a joint effort with members from Inria, the prototype will allow to show our ideas about dissemination of data under flash crowds conditions and verify some of the simulation done by assaf and yotem.
Assaf researched the problem of distributing a large quantity of packages to a heterogeneous community of users. For example, when a new GNU/Linux update becomes available, each user will need a different subset of packages according to her specific HW/SW configuration. Using popular tools such as BitTorrent is problematic since it forces each user to either download the whole update distribution (might be very large and mostly irrelevant) or download specific packages (inefficient when the number of required packages is large). The research suggested a novel method of dividing the users into several sampling groups, such that the distribution within each sampling group is efficient. Although the problem itself was proved to be NP-Hard, approximation algorithms were proposed and shown in practicle (using an adjusted BitTorrent simulator) to be close to the optimal solution.
Yotam is currently finalizing his thesis, which tackles the topic of Free and Open Source Software distribution with the main focus being the efficient and cooperative download of new releases of Linux distributions. The thesis presents different clustering algorithms for solving the problem and shows both theoretical and experimental results using both synthetic and real-life inputs as the benchmark.