Think distributed : a glimpse of decentralized algorithmic ∗

Since the birth of the computer era, the clientserver paradigm has become the natural way of designing network architectures. As data and application logic are available at just one or few places (the servers), obvious simplicity follows concerning deployment, security or maintenance questions. Large mainframes and server farms are still nowadays at the core of all critical applications in industry or in public domain. Only recently occurred one of the first original breakthrough w.r.t. this almighty monolithic paradigm; Napster and Freenet proposed around 1999 a new concept called peer-to-peer, were the client also plays the role of a server for the community of users. Those two states played simultaneously by users’ computers relaxed the need for a powerful central server (in the case of Napster), or even removed the necessity of the server existence itself (case of Freenet). Many applications followed, from IP telephony, media streaming, virtual worlds, mail, to chat/messengers. Amongst properties that distribution allows, we can cite: (i) a very large number of participants (nowadays in millions), that centralized servers are unable to handle or at a prohibitive cost, (ii) fault tolerance and resilience to attacks (no single point of failure), (iii) anonymity (Freenet, ANTs P2P...), as the protocol’s traffic do not transit through a single entity, and (iiii) computing resource exhaustive tasks, impossible to compute in reasonable time even by supercomputers (e.g. protein folding). Distributed systems include, along with peer-to-peer approaches, grid computing, sensor networks, and at a lower level multi-processors or distributed shared-memories. At the application level, software algorithmic must be rethought. This is a direct consequence of the shift from a single “all seeing” point, to the distribution of program logic to a crowd of computers. The relative ease of algorithmic given by the fact that all data or client computers have a rendez vous at a particular place, is considerably complexified by the distribution of tasks.