Reinforcement Learning-Based Service-Oriented Dynamic Multipath Routing in SDN

Figure 16

Performance of three path distribution schemes in the realistic six-service scenario (the average values are calculated from the 500th to the 1500th second after convergence). (a) Reward gained by SPD; (b) reward gained by DQN; (c) reward gained by RED-STAR; (d) average reward of three services of each scheme; (e) maximum bandwidth utilization of each scheme.

(a)

(b)

(c)

(d)

(e)