Intelligent CAC and routing for multi-point connections
Tien Pham Van, Franz Josef Rammig, Yoshiaki Tanaka · International Conference on Communications · 2004
This study introduces Reinforcement Learning (RL) to solve the problem of Call Admission Control (CAC)&Routing for multi-point connections in networks serving multiple service classes. The network system is trained to find out the optimal control policy which brings up the highest amount of reward in longrun. For a manageable solution and realizable training time, decomposition of network and connections into link level is implemented. To demonstrate the prominence of RL-based routing against MOSPF (Multicast extension to Open Shortest Path First) protocol, a routing protocol with high performance among available ones, we consider different criteria, including reward rate, call drop rate, and link usage rate.