Design And Implementation Of An Intelligent Tuning System For Separated Computing And Storage Database Cache

In a separated computing and storage database,the input and output(IO)bottleneck of the database turns to the network IO between the computing node and the storage node and the disk IO of the storage node.Cache technology is used to balance the processing speed gap between CPU and IO.Because different loads have different requirements for the cache,the static cache configuration will affect performance and waste resources under different loads.Therefore,it is necessary to configure the parameters of the cache of each node in the separated computing and storage database according to different loads,so as to improve the ability of nodes to provide services on the basis of rational use of resourcesThrough the analysis of the realization principle of the separated computing and storage database and the working mechanism of the cache in each node,a scheme for intelligently tuning the cache of each node under the separated computing and storage database is designed.The solution uses reinforcement learning methods to make it possible to analyze the state of the buffer pool to give appropriate cache parameter recommendations when facing different loads.The state of buffer pool is determined by 18 monitoring parameters in four aspects: physical state,state of IO,state of least recently used(LRU)list and state of activity;The cache parameters recommended to computing nodes include 5 parameters to be adjusted in three aspects: cache capacity,cache management strategy and page elimination mechanism;The cache parameters recommended to storage nodes include 11 parameters to be adjusted in four aspects: cache capacity,cache management strategy,page elimination mechanism and page flushing mechanism.Based on the analysis of the existing reinforcement learning algorithm and cache tuning environment,the depth deterministic strategy gradient algorithm is selected for parameter recommendation.According to the tuning objectives of improving performance and saving resources,the reward function with cache hit rate and cache resource allocation as the main body is determined.The implemented system includes a monitoring data acquisition module,a parameter recommendation module,and a parameter adjustment execution module.Experiments are designed to verify the effectiveness of the recommended parameters in the cache intelligent tuning system.The experimental results show that compared with the default parameters,the performance of the parameters recommended by the cache intelligent tuning system will be improved in the face of different loads,and the performance is close to or even better than that of the parameters with the largest resource allocation,so as to achieve the effect of cache adapting to the load,and verify the effectiveness of the cache intelligent tuning system.