Dynamic load balancing with learning model for Sudoku solving system

This paper proposes a dynamic load balancing with learning model for a Sudoku problem solving system that has multiple workers and multiple solvers. The objective is to minimise the total processing time of problem solving. Our load balancing with learning model distributes each Sudoku problem to an appropriate pair of worker and solver when it is received by the system. The information of the estimated solution time for a specific number of given input values, the estimated finishing time of each worker, and the idle status of each worker are used to determine the worker-solver pairs. In addition, the proposed system can estimate the waiting period for each problem. Test results show that the system has shorter processing time than conventional alternatives.