The Optimization Of Persistent Caching Mechanism Based On The LSM-Tree Model

The booming development of internet led to an explosive growth of data quantity. Traditional relational database cannot meet the requirements of the massive data-scale applications, so the key-value(KV) storage mechanism is proposed. NoSQL database, which is based on this mechanism, is widely promoted because of its fast superior performance in processing massive amounts of data and thus is generally used as front end of traditional relational database. However, KV database is memory-based and thus shows defect in data persistence, which results in possibility of data loss when the system is down. However, fast development of internet technology brings new ideas of caching data into persistence in KV database.This thesis studies the persistence mechanism of KV storage, which is widely used in NoSQL databases. The LSM-Tree persistence mechanism in Google's open source database LevelDB is deeply studied. Firstly, we analyze the LSM-Tree persistent cache model and its caching mechanism. Secondly, modified LSM-Tree model is proposed by complementing the shortcoming of the LSM-Tree cache mechanism that it induces slower random read throughput although the write throughput is very fast. Thirdly, specific attention is paid to the application of the stand-alone caching mechanism of LevelDB. The key technical and operational mechanism of LevelDB is further modified using C ++ language.Lastly, we use Yahoo's YCSB load generator framework to compare the performances of LevelDB before and after the modification. The results show that read throughput doubled after our modification with the same write throughput and random read delay is significantly reduced. The modification and optimization of LevelDB show superior performances.