The Research On The Optimized Placement Strategy And Fault Tolerant Storage In Data Center Network

With the rapid development of cloud computing,the data center as the core infrastructure of cloud computing has also begun to grow rapidly.The centralized network distribution of traditional data centers is also broken.With the frame of distributed instead of centralized,today's data centers place high demands on efficient storage and management of massive data,a large number of communication modes with frequent changes and low round-trip delay,etc.The traditional protocols and algorithms are difficult to contribute in the existing data center networks.In recent years,a great deal of research has focused on how to optimize the data placement and storage strateg ies in data centers.These researches mainly include two aspects: first,how to implement an efficient content placement strategy and choose the optimal storage node,reducing access latency while improving load balancing.Second,how to implement a fault-tolerant storage strategy to reduce the data r ecovery time after node failure.Minimize the impact of node failure.Reinforcement learning algorithm is a kind of algorithm related to dynamic programming,wh ich is suitable for solving the optimization problems of data placement.In this paper,according to the application of reinforcement learning,we proposed the optimized placement strategy based on reinforcement learning algorithm.Based on the erasure code technique,a fault-tolerant storage strategy with data association method is also proposed.According to the strategy of content placement in data center,the first part of this paper applies the Q-learning method of reinforcement learning to content placement problems in data center network.Taking the average link bandwidth and storage load as consideration,this paper proposed a Q-learning method based content placement strategy.This topic uses a Fat-tree data center network topology that is widely used in switch-centric architectures.The placement algorithm mainly considers the average link bandwidth,the data access delay and the load of storage nodes,and establishes the corresponding joint function as the standard of the evaluation for placement strategy.Through comparison experiments,we find the placement strategy can effectively reduce data access latency and achieve optimal load balancing,which is because of balanced link condition and node load.Then in the second part of this paper,a fault-tolerant storage strategy based on the balance of encoded data is proposed.The existing placement strategies are mainly based on the redundancy and the access time.In order to solve the data failure problem caused by the association of encoded data,we proposed a balanced classification storage strategy.The experiment shows that the placement strategy can effectively reduce the data failure probability and effectively decrease the data access latency and Mean Time to Failure(MTTF).