Toward efficient resource utilization at edge nodes in federated learning