There are lots of multimedia services such as YouTube and Netflix in the Internet. The multimedia services through the Internet will continue to grow. Most of these multimedia services generally stream a content to many users. In these type services, the multicast transmission mode can efficiently deliver the content to multiple subscribers. Especially in SDN (Software-Defined Network), multicast mode can be easily adopted because the centralized controller sets up all routes using multicast tree with global network information. However, the construction of multicast tree is an NP (Non-deterministic Polynomial-time) problem and is hard to make the optimal multicast tree in the real world. Therefore, in this paper, we propose a heuristic way to generate a multicast tree using DQN (Deep-Q-Network) which is a type of reinforcement learning in machine learning field. Through the experiment, we show that the performance ratio of the proposed algorithm is 1.21 with the topology of 10 nodes and it generates the multicast tree better than the previous heuristic algorithms such as TM (Takahashi and Matsuyama) algorithm.
展开▼
