Gossip-based self-organization19/01/2007 au 20/01/2007
Intervenant(s) : Spyros Voulgaris (ETH Zurich)
Recent years' emergence of massive-scale distributed applications (in the order of 10K or 100K nodes) poses challenging issues regarding their management. Traditional centralized or deterministic protocols are often inadequate or involve high administration and infrastructure costs to handle systems of such scale, notably when nodes regularly join and leave the system (either purposefully or on account of failures) and when reliability should not depend on the availability of any particular node.
Gossiping protocols form an attractive alternative. They offer strong reliability guarantees on a probabilistic basis, enjoy remarkable resilience to failures, demonstrate self-healing behavior, and are inherently scalable, while retaining a simple and inexpensive character. The topology governing the communication between nodes is crucial to the aforementioned properties.
In this talk I will present my research on fundamental gossiping protocols that enable networks to self-organize in the following two classes of topologies:
(A) Randomized topologies. First, they strive at maintaining the network connected in the face of massive failures or high churn. Second, they enable communication between random nodes, essential to information dissemination, resource discovery, etc.
(B) Structured topologies. The network self-organizes to a structure depending on the distributed application in question (searching, routing, monitoring, dissemination, pub/sub, etc.)
I will, finally, demonstrate the applicability of my protocols by combining them to build Sub-2-Sub, a fully decentralized, autonomous, self-organizing, administration-free, attribute-based Publish/Subscribe system that supports exact and range subscriptions.
Spyros Voulgaris is a postdoc researcher in the Department of Computer Science at the Swiss Federal Institute of Technology in Zurich , working in the Information and Communication Systems Research Group led by professor Gustavo Alonso. He obtained his Ph.D. in 2006 from the Vrije Universiteit in Amsterdam, under the supervision of professors Maarten van Steen and Andrewv S. Tanenbaum. His Ph.D. thesis title is "Epidemic-based Self-Organization in Peer-to-Peer Systems". He also holds a M.Sc. form the University of Michigan in Ann Arbor (1999), and a B.Sc. from the University of Patras, Greece (1997). He has worked for two years (1999-2001) for Hughes Network Systems in Maryland, USA, on developing network management applications. He has also done research internships at Microsoft Research (Cambridge, UK, 2003), and Hewlett Packard Labs (Palo Alto, CA, USA, 1998). His current research interests are in the area of very large scale decentralized systems.