This thesis focuses on resource allocation challenges in 6G-based IoT networks under practical constraints. We begin by addressing infeasibility in centralized systems, where limited radio resources make it impossible to serve all users' reliability requirements. A robust scheme is proposed to maximize the number of satisfied users under such infeasible conditions. Next, we explore a distributed solution for cell-free massive MIMO networks to overcome computational bottlenecks. A graph neural network (GNN)-based framework is developed, allowing access points to make local decisions without centralized coordination. Finally, we integrate both approaches to solve infeasible allocation problems in a distributed cell-free massive MIMO setting. The resulting framework enhances scalability, reduces congestion, and improves reliability for next-generation 6G IoT systems.