Speculative Nondeterminism

We propose a new programmable concurrency control framework called speculative nondeterminism for real time, “open” and distributed agents. In this framework, dynamic and concurrent agent programs affect each other through a centralized store which represents shared resources. Our framework has novel programming constructs which allows an agent to speculate against the future by multiple exclusive choices that encode different strategies to achieve its goals. The agent can make multiple choices which execute together but in isolation, thus, there is a combinatorial effect through the joint interaction. Unlike other concurrency frameworks, speculation models many potential outcomes and thus improves the chances of an agent reaching a desired outcome. Each possibility is represented by a “virtual world.” Agents operate in the virtual worlds created by the system, the framework provides mechanisms for agents which have reached their objective but yet are represented by multiple possible worlds to cleanly exit. We also provide mechanisms to reduce the possibilities where an agent can commit to certainly possibilities and give up on others, but this is rather different from conventional committed choice. This paper presents the design, operational semantics of speculative nondeterminism, small but practical examples illustrating the use of the language and experimental results of a prototype.