This study explores accelerating strategies in column generation (CG) to effectively solve online dial-a-ride problems in large-scale ride-sharing systems. Traditional heuristics often lack guarantees on solution quality, and exact methods like CG can be computationally intensive for real-time applications. To address these challenges, we introduce and outline strategies aimed at speeding up the subproblem-solving phase during the solution process in real-time contexts. These subproblems are typically formulated as the Shortest Path Problem with Resource Constraints, in a CG approach, and are traditionally solved via dynamic programming. Our methods significantly enhance the scalability of CG in real-time contexts. We validate the approach using historical taxi trip data from New York City, handling up to 30,000 requests per hour. Computational experiments demonstrate significant reductions in processing times and the ability to produce high-quality solutions more efficiently.