Section: Application Domains
Robust and Distributed Algorithms, Algorithmic Game Theory
One of the problems related to distributed algorithmics corresponds to the minimization of resources (time of transit, quality of services) in problems of transiting information (routing problems, group telecommunications) in telecommunication networks.
Each type of network gives rise to natural constraints on models. For example, a network is generally modeled by a graph. The material and physical constraints on each component of the network (routers, communication media, topology, etcÂ ...) result in different models. One natural objective is then to build algorithms to solve those types of problems on various models. One can also constrain solutions to offer certain guarantees: for example the property of selfstabilization, which expresses that the system must end in a correct state whatever its initial state is; or certain guarantees of robustness: even in the presence of a small proportion of Byzantine actors, the final result will remain correct; even in the presence of rational actors with divergent interests, the final result will remain acceptable.
Algorithms of traditional distributed algorithmics were designed with the strong assumption that the interest of each actor does not differ from the interest of the group. For example, in a routing problem, classical distributed algorithms do not take into account the economic interests of the various autonomous systems, and only try to minimize criteria such as shortest distances, completely ignoring the economical consequences of decisions for involved agents.
If one wants to have more realistic models, and take into account the way the different agents behave, one gets more complex models.
However, today, one gets models which are hard to analyse. For example,

Models of dynamism are missing: e.g., how to model a negotiation in a distributed auction mechanism for the access to a telecommunications service,

only few methods are known to guarantee that the equilibrium reached by such systems remains in some domains that could be qualified as safe or reasonable,

there is almost no method discussing the speed of convergence, when there is convergence,

only a little is known about the time and space resources necessary to establish some techniques to guarantee correct behavior.
Thus, it is important to reconsider the algorithms of the theory of distributed algorithmics, under the angle of the competitive interests that involved agents can have (Adversary computation). This requires to include/understand well how to reason on these types of models.