sábado, octubre 28, 2006

El futuro del diseño, en veinte años

Volviendo a Lambda, the ultimate, Peter Van Roy escribe dos líneas sobre el futuro del diseño, sugiriendo vías de evolución, y proponiendo un modelo:
What will programming look like in 20 years? Maybe it will be based on a "definitive language" like the speculations of the article Convergence in language design: a case of lightning striking four times in the same place (FLOPS 2006). Such a language will have a layered structure and its handling of concurrency is important. (Whether it is dynamically or statically typed is of absolutely no importance, BTW.)

How will we program with such a language? Maybe we will program with feedback loops, as explained in Self management and the future of software design (FACS 06). This seems to be one way to handle complexity, the inevitability of software and hardware faults, and managing global behavior of a system. I am coordinating a new project, SELFMAN, that is looking at this.

Extraído del papel de Van Roy mencionado:
We propose to build self-managing software systems as sets of concurrent agents interacting by means of asynchronous events and implemented using a component model with first-class
components and component instances. In this framework, self-managing systems are built as hierarchies of interacting feedback loops. The first design rule is that the whole system (except perhaps a small kernel) should be inside a feedback loop.
Feedback loops interact through two mechanisms, stigmergy (shared environment parameters) or management (one loop controls another). The feedback loop structure is designed to provide a desired global behavior. This behavior should also be predictable from the loop structure. We relate this proposal to two other architectures, namely the Erlang fault-tolerance architecture and the subsumption architecture for implementing intelligent behavior.
These ideas are being realized in SELFMAN, a project in the European 6th Framework Programme that started in June 2006 [27]. We intend to elaborate these ideas into a programming methodology together with an implementation. It should be as easy to program with and reason about a feedback loop as it is for an object or a component. We will design and formalize a component model that is based on the Oz kernel language extended with elements from the Fractal model.
We will use this component model as the basis of a programming model along the lines of Section 5 and implement this model in Mozart [26,9,13,23]. We will build a feedback loop architecture on top of this implementation and use it to implement a self-managing replicated transactional storage service.
Otros puntos de vista en la misma línea pueden verse en Paradigmas.

No hay comentarios.: