Affecting Factors Of Database Management System Performance Based On Storage-side Data Processing Model

In a traditional database management system,all data processing is done in the host,and a large amount of data needs to be transferred from the external storage device to the host,so the data transfer process is one of the bottlenecks that affects the performance of the system.With the rapid development of smart storage devices(such as Smart SSD,Smart Solid State Drive),it has certain computing power and high bandwidth internally to meet the needs of storage-side data processing models.This model can solve this performance bottleneck well.The core idea of the storage-side data processing model is to decentralize some calculations to run in intelligent storage devices,preprocess the data,and reduce the amount of data transmitted on the bus.However,the current data calculation capability of solid-state drives is weaker than the central processor in the host,and the decentralized calculation needs to be screened to ensure the optimization effect of the model.Therefore,in the storage-side data processing model,in order to effectively improve system performance,the screening work of decentralized computing under various influencing factors is an important research topic.However,from the current research,the existing research lacks a comprehensive analysis model of the performance influencing factors of data processing on the storage side,it is difficult to ensure the effect of each optimization,and it is impossible to scientifically guide the system performance optimization work.In view of this,the research work of this paper is as follows:1)Research on DBMS data processing and performance influencing factors: First,on the basis of traditional system architecture and storage data processing architecture,analyze the difference between storage data processing model and traditional data processing model,and then identify the data migration and data processing process The main performance overheadpart;then compare the data processing overhead formulas of the two models in detail,analyze the similarities and differences in the data processing process;and finally combine the traditional DBMS ideas with the working principle of the storage-side data processing model to summarize the impact of the storage-side data processing model Three aspects of system performance improvement.2)Research on the influencing factors of the pre-relationship of database operators: In view of the pre-execution relationship found among the database operators in the research,if the decentralized operators are improper,it will cause the problem of reverse data transmission,which will lead to a further increase in the amount of data transmission on the bus and a negative increase in optimization.Then,based on the operator pre-relationship,a pre-relationship probability table is proposed to prioritize the decentralization of operators.Finally,combining the pre-relationship and predecessor scheme,a decentralized operator screening scheme based on pre-relationship is proposed,and the superiority of the screening scheme is compared through experiments.3)Research on influencing factors of data characteristics: Data characteristics refer to various parameters involved in data storage based on DBMS,such as database table size.Data characteristics will directly affect the query speed of target data and the processing speed of target data in the query process.This chapter mainly analyzes the three main aspects of the size of the database table,the filtering ratio of the database table,and the index structure of the database table.Performance estimation methods and performance change trends under influencing factors.The final experiment shows that the impact of these three data characteristics on system performance is not linear;there is a critical value.Only when the critical value is reached,the optimization effect will be gradually highlighted,and the size of the critical value is affected by many aspects,and will change with the change of the environment.4)Research on factors influencing device performance: System deviceperformance refers to the operating parameters of system hardware,such as the CPU main frequency and system bandwidth.Since CPU resources are directly related to data processing speed,and I / O resources are directly related to data transmission speed,both of these resources are likely to cause system performance bottlenecks.Therefore,the optimization of these two aspects is very conducive to the improvement of system performance.This chapter gives performance estimation methods and performance change trends under different influencing factors,and experiments show that optimization of equipment performance influencing factors has limited improvement in system performance.There is a critical value for the optimization effect,and it will not continue to grow with the improvement of the performance of the device.As long as the critical value is reached,there is no need to optimize the factors affecting the performance of the device.In summary,this paper mainly studies the performance influencing factors under the data processing model of the storage side from theoretical analysis and experimental verification.First,theoretically analyze the similarities and differences between the corresponding data processing models through the traditional system architecture and the storage-side data processing system architecture,and summarize the main factors that affect system performance based on the difference in data processing overhead between the two.The experiments show that the optimization effect of the screening scheme based on the pre-relationship under all influencing factors is significantly better than the screening scheme based on the ratio of input /output data volume,and the performance estimation under different influencing factors is given.The trend of method and performance changes,and the optimization effects of different system architectures are compared,and the corresponding evaluation is made according to the experimental results to guide optimization of the DBMS under various influencing factors.