The k-server problem: a reinforcement learning approach

This work presents an original algorithm, based on reinforcement learning, as a solution for the k-server problem. This problem was modelled as a multi-stage decision-making process, and subsequently the Q-learning algorithm was used as a means of solution. A series of experiments was conducted with the aim of evaluating the appropriateness and performance of the proposed solution. The results obtained show the efficiency of the suggested algorithm in comparison with other methods of solving the k-server problem frequently cited in the literature, whose rates of competitiveness have already been proven.