MASP-Reduce: A Proposal for Distributed Computation of Stable Models

There has been an increasing interest in recent years towards the development of efficient solvers for Answer Set Programming (ASP) and towards the application of ASP to solve increasing more challenging problems. In particular, several recent efforts have explored the issue of scalability of ASP solvers when addressing the challenges caused by the need to ground the program before resolution. This paper offers an alternative solution to this challenge, focused on the use of distributed programming techniques to reason about ASP programs whose grounding would be prohibitive for mainstream ASP solvers. The work builds on a proposal of a characterization of answer set solving as a form of non-standard graph coloring. The paper expands this characterization to include syntactic extensions used in modern ASP (e.g., choice rules, weight constraints). We present an implementation of the solver using a distributed programming framework specifically designed to manipulate very large graphs, as provided by Apache Spark, which in turn builds on the MapReduce programming framework. Finally, we provide a few preliminary results obtained from the first prototype implementation of this approach.