The Research And Implementation Of Memory Management Mechanism For Cassandra

Cassandra database system is one of the NoSQL system.It was developed by Facebook and finally became one of the Apache top object after being an open source system.Nowadays,Cassandra distribute NoSQL system has been the Top10 of the database rankings.Because of its good scalability,it has been adopted by Digg,Twitter and famous Web2.0 site.During the actual operation and maintenance process,it can often achieve a higher response written rate.But because of the big data redundancy caused by the writing mechanism,it will influence the reading rate.The combination mechanism realized by the system can alleviate the impact to the reading rate due to data redundancy.But the unreasonable consolidation strategy and timing will also affect the performance of the database.What's more,the different allocation strategy of the memory space will also affect the performance of every aspect of the system.The main research of the article is how to improve the unreasonable method of the combination mechanism.To solve this problem,this article puts forward a dynamic data management system to improve the combination mechanism and increase the success rate of the combination process.At the same time,this article also establish a relationship model between the different memory space and the system performance to guide the actual running application maintenance.Our main work can be summarize as follows:1)We analyses the reading and writing process of the Cassandra database system.According to the conclusion that it has an imbalance reading and writing performance and especially a bad reading performance,we put forward a more efficiency dynamic data management mechanism based on the former combination mechanism and application interface.2)We also improve the dynamic data management mechanism by choosing more reasonable file,file prefetching,combination with multi-threads and system surveillance to improve the performance of the dynamic data management mechanism.The test result shows that this mechanism can accelerate the combination process to 15% compared to the former performance and the speedup of the reading request can reach about 2.3)The memory space of the Cassandra database system can be divided into different function unit and both of them can be adjusted by the parameters.Actually,we always allocate the memory space according to experience and repeated attempts.But it won't give us the best answer for the allocation of memory space.This article analyses the relationship between the different functional units and the system performance.4)We first establish the functional relationship between the functional units and system performance.Then we establish a quantitative model of the reading and writing performance by tests.We have already realized the relationship model for real database servers to guide memory allocation and performance evaluation in practical application.The simulation results show that this memory model can describe the quantization relationship of the memory space and system performance commendably.