Intelligent packet scheduler for general packet radio service

The general packet radio service (GPRS) augments GSM to provide packet switched data services to mobile users. Packet scheduling in GPRS is dynamic and several scheduling techniques have been implemented, for example round robin; however, these generally assure only best effort quality of service. We compare prominent scheduling algorithms by simulation of Web and E-mail traffic finding that earliest deadline first and first come first served scheduling perform well with few users, but round robin is preferable with large numbers of users. We introduce a novel scheduling algorithm, based on reinforcement learning, for scheduling packets according to quality of service. Simulation studies show that it outperforms a naive prioritised round robin algorithm and can adapt to changing network conditions.