Talk Amongst Yourselves: Gossip as an Introduction to Self-Organizing Algorithms

This is a past event

27 people went

Details

Info about the venue:

* check in at the front desk
* meeting will be on the 18th floor
* doors lock at six, so go around to the back of the building if you arrive late, and security will let you in.

Paper Title:

Márk Jelasity, Rachid Guerraoui, Anne-Marie Kermarrec, and Maarten Van Steen. "The peer sampling service: Experimental evaluation of unstructured gossip-based implementations." In ACM/IFIP/USENIX International Conference on Distributed Systems Platforms and Open Distributed Processing, pp. 79-98. Springer, Berlin, Heidelberg, 2004.

Paper Link:
http://www.inf.u-szeged.hu/~jelasity/cikkek/middleware04.pdf

Talk Abstract:
Self-organizing algorithms enable the design of fully decentralized computing systems with useful properties that emulate those found in nature, such as adapting to changing conditions or self-healing when damaged. Many metaphor-driven models have been built in this space, drawing inspiration from fields such as biology, physics, chemistry, ecology, economics, social system.

A simple example of a self-organizing mechanism applied to distributed systems can be found in "gossip protocols", which use dynamics inspired by observations of the spread of information through social networks. As such, they serve as a useful entry point to thinking about self-organization as a tool for designing real-world computing systems. Gossip protocols are also useful in their own right, and are used by NoSQL databases such as Cassandra and blockchain implementations such as Bitcoin and Ethereum.

In this session, we'll take a look at the design and implementation of Newcast gossip protocol, which is described in a number of papers. The selected paper discusses applying Newscast to the problem of peer sampling, which ensures that every participant in a leaderless network has a fresh selection of other peers with whom to exchange information.