A Distributed Breakout Algorithm for Solving a Large-Scale Project Scheduling Problem

This paper describes an application, an algorithm and the modeling of a distributed CSP. A real-life, distributed, resource constrained, large-scale, project-scheduling problem is introduced. This problem is modeled and formalized as a distributed CSP. It is discussed and demonstrated how local, heterogeneous and ambiguous project coordination information of the project partners can be exchanged unambiguously by using a project planning ontology and by translating coordination information into domain variable values and constraints. Moreover a coordination method is presented which is designed towards solving the large-scale problem. This coordination method includes two separate algorithms. The first algorithm is a distributed forward pass, which computes an initial variable value assignment in polynomial time. The initial variable value assignment is then used as the entry point for the second algorithm. The second algorithm is a distributed breakout algorithm, which iteratively improves the initial variable value assignment with a conflict-minimizing strategy until a solution is obtained. For improving the search efficiency of the breakout algorithm two techniques, domain filtering and single sided precedence constraints, are implemented. The overall coordination algorithm as well as the efficiency techniques are tested and evaluated with three real sub-problems as well as a set of random problems.