Over the last ten years, distributed systems have undergone a considerable change of scale. Nowadays, through local and wide area networks (LANs and WANs), a very large number of computers and microcomputers are connected across the world. Which, if we include mobile telephones, notebooks, PDAs and other computing-capacity enabled devices, means tens of millions of machines which can, potentially, talk to each other across intermediate cable, infra-red or radio relay networks. If we add to this the constant increase in the volume of multimedia data being exchanged at any given time worldwide, we gain a measure of the complexity involved in defining and implementing protocols capable of managing all the machines in a large distributed system with, additionally, a very high level of performance and reliability.
There has been a paradigm shift with the emergence of large-scale dynamic distributed systems," observes Anne-Marie Kermarrec. There is now a high degree of uncertainty in the operation of these giant networks that now needs to be taken into account and controlled, and in an entirely decentralised way. In order to meet these needs, one of the major challenges is to design protocols enabling a large distributed system to manage, organise and repair itself without any one entity in the system having complete knowledge. Achieving a position where each entity's individual decisions, based on restricted knowledge of the system, emerge from global properties represents one of our major challenges.
Rumours and bacteria
To achieve this objective, it is necessary to take account of the theoretical and practical aspects of large-scale distributed computing. Analogous operations have been found not only in epidemic protocols (multiplication and spread of bacteria) but also in the process by which a rumour is propagated within a given population. The ASAP project team put them to good use to show that, if each element in a set "knows" and passes on a piece of information to a very small number of other elements chosen at random, the information spreads reliably and quickly. This same analogy can be used to create and maintain large networks, perform distributed calculations and even navigate simply and efficiently around a large-scale distributed world. Begun in 2006, the ASAP project team's research is organised around three topics: distributed calculation models, the management of various types of resource (computing, data, events, bandwidth, etc.) and data collection and propagation in sensor networks. More recently, the Gossple project aims to provide an entirely decentralised approach to web navigation, based on epidemic protocols, to exploit preferences and tastes by placing the user at the heart of the search process.
Anne-Marie Kermarrec, awarded the ERC Starting Grant “Gossple: a radically new approach to navigating the digital information universe”
Anne-Marie Kermarrec, ASAP project team leader, Inria Rennes - Bretagne Atlantique centre
- Since 2004: Research director, Inria Rennes. ASAP is a joint project team with the French National Centre for Scientific Research - CNRS, Rennes I University and the French National Institute of Applied Sciences - INSA, Rennes Dividing her time between Rennes and Saclay, she instigated the cooperation with Vrije University in Amsterdam within the Epi-nets associate team, conducting in-depth research in the field of unstructured epidemic network applications.
- 2000 / 2004: Researcher at Microsoft Research in Cambridge (United Kingdom).
- 1997 / 1999: Senior lecturer at Rennes University.
- 1996 / 1997: Post-doctoral researcher at Vrije University, Amsterdam (Netherlands).