Multi-objective Disaster Backup in Inter-datacenter Using Reinforcement Learning

With rapid growth of data centers and great rising concern on data security in recent years, disaster backup has attracted the attention of many researchers. Most researchers focus on reducing the backup bandwidth cost by using multicast routing, but pay less attention to load balance. The local link congestion will seriously affect the user's experience interacting with the data center. To optimize bandwidth cost and load balance simultaneously, we use multicast routing and store-and-forward mechanism to build multiple disaster-backup multicast trees, and forward data at appropriate time slots to reduce link congestion. To solve the weight selection problem between cost and load balance, we propose the multicasting backup multi-objective reinforcement learning algorithm based on Chebyshev scalarization function, which significantly improve the hypervolume of solution set. Simulation results show that our algorithm outperforms existing algorithms in terms of bandwidth cost, maximal link congestion and the hypervolume of solution set.