Research On Caching-based Key-value Storage System Performance Optimization Scheme

In the context of the information age,while people enjoy the convenience of digital technology,they are also facing the problem of large-scale data interaction.Key-value store databases represented by Level DB are widely used in write-intensive scenarios because of their excellent write performance.However,in scenarios with high requirements for read performance,it is often inadequate,especially in the environment of large data.Multiple accesses to duplicate data consume a lot of CPU resources and incur unnecessary overhead,which can also cause damage to system performance.Caching is an effective measure to address the additional data overhead caused by duplicate data.The read performance of the key-value storage system can be improved by introducing caching,such as the four cache mechanisms of cache traversal,cache avalanche,cache breakdown and cache warm-up adopted in Redis,and the LRUCache cache mechanism introduced in Level DB.To further enhance the performance gains that caching brings to key-value storage systems,this paper has done a lot of work on the basis of the LRUCache caching mechanism of Level DB,a key-value storage system based on LSM-Tree.It aims to improve system performance by modifying the underlying cache structure and cache algorithm.The main research contents and optimization schemes are as follows:(1)Proposing a caching scheme for multi-hash tables.By increasing the number of hash tables,this solution optimizes the layout of points in the cache,postpones the execution of rehash operations caused by too many nodes in a hash bucket and improves the read performance of the system without affecting the write performance.(2)Proposing to apply the MRU cache replacement strategy to the cache system of Level DB.The new optimization scheme abandons the LRU cache replacement strategy adopted by the original LRUCache cache system and alleviates the problem of slow read performance caused by the incoming duplicate old data to a certain extent.Experimental results suggest that the two optimization schemes presented in this article bring certain improvements to the read performance of the system without affecting the write performance.Increasing the hash table scheme can improve the read performance of the system by 5% and the MRU cache replacement strategy can improve the read performance of the system by 5%～25% because of its applicability to duplicate data.