An Adaptive Migration Strategy Based On Redis

In recent years,with the rapid development of information technology,such as cloud computing,Internet of things and social networks,the amount of data generated by various industries is growing at an unprecedented rate,and a new era of big data is coming to us.The NoSQL database represented by Redis is praised by the industry for its easy extension and high performance.The so-called extensibility of Redis includes two aspects: On the one hand,it means that Redis can support the storage of a great deal of data.Its distributed architecture makes it possible to store more data as long as there are more machines;On the other hand,it can support a large number of concurrent queries,which process faster.The so-called availability of Redis means that once the system fails,it can still work through failover.Redis achieves scalability and availability at the expense of strong consistency.Therefore,it is particularly important to optimize the scalability and availability of Redis.Migration is the key technology to realize the dynamic expansion of node and elastic load balancing.In order to improve expansibility,this paper presents an adaptive migration strategy of unit maximum benefit.On one hand,the strategy solves the shortage of manual migration of Redis by judging load of cluster system real-time and achieves adaptive control between systems by strategic migration,which leads to static storage equilibrium.On the other hand,the strategy proposes a migration scheme of unit maximum benefit to optimize migration plan.The scheme will choose the optimal node and find the best strategy until all migration are finished.Experimental results show that migration time,CPU usage and throughput are optimized.The surge of hot data access results in a huge pressure in some nodes.This paper,combining Master-Slave mode and Replica Migration strategy of Redis and MQ algorithm,proposes an adaptive migration strategy for hot and cold duplicate.The strategy can reduce the pressure of high frequency access of hot data by increasing the redundancy of the master node of hot data.Adding a new slave duplicate is too costly,and this paper borrows slave duplicates from other master nodes and these slave duplicates come from the master node of cold data according to the idea of Replica Migration strategy.The access frequency of the cold data is low,and reducing the slave duplicate of cold data does not affect the availability of cluster system as a whole.Experiments show that this strategy can achieve higher hit rate,less response time and higher throughput.