Predictable Self-Organization with Computational Fields

Abstract

In recent years, a number of different strands of research on self-organizing systems have come together to create a new "aggregate programming" approach to the engineering of distributed systems. Aggregate programming is motivated by a desire to avoid the notoriously intractable "local to global" problem, where the system designer must predict how to control individual devices to achieve a collective goal. Instead, the designer programs an abstraction of the collective, composing "building block" primitives from a library of special cases where the local-to-global problem is already solved. Unifying a number of the proposed aggregate programming approaches is the notion of a "computational field" that maps each device in the field's domain to a local value in its range. This concept was originally developed for spatial computers, in which communication and geometric position are closely linked, but can support effective aggregate programming of many non-spatial networks as well. A mathematical foundation for such approaches has been formalized recently with a minimal "field calculus" that appears to be an effective unifying model, covering a wide range of aggregate programming models, both continuous (e.g., geometry-based) and discrete (e.g., graph-based). On this foundation, restricted languages can ensure various desirable properties such as scalability, self-stabilization, and robustness to perturbation. By building up a sufficiently broad collection of composable "building block" distributed algorithms, it is possible to enable simple and rapid development of complex distributed systems that are implicitly scalable and resilient. The ultimate aim of this line of research is to make the programming of robust distributed systems as simple and widespread as single-processor programming, thereby enabling widespread increases in the reliability, efficiency, and democracy of our technological infrastructure.