Distributed search is important for finding solutions to hard problems in artificial intelligence. Building distributed search systems can be difficult because the steps required to solve these problems are interdependent. Fortunately, aspects of search systems exhibit commonalities that allow them to be distributed using several different paradigms. These paradigms can be used as the basis for libraries to make implementing new systems, or distributing existing systems, easier. DisSLib:CC is such a library.
DisSLib:CC implements the distributed search paradigm search with a central common search state. In this paradigm, agents collaborate to update a central search state. Systems built with DisSLib:CC require very little extra code to implement compared to the standalone versions. These systems can also show improvements in wall-clock run-times, which can be improved by varying the meta-parameters of the distribution paradigm. One such parameter is the number of steps, transitions, that the agents execute before consulting the common search state and in this paper we show how varying this meta-parameter improves the efficiency of the systems.
Generated: 11/12/2014