Three-Layout Key/Value Index System Base On JB~c-LSM

With the rapid development of Internet technology, the quantity of network access data grows exponentially. It is urgent to improve the management ability and the access abili-ty of huge amounts of data. JBc-LSM system is developed by ATR Lab to handle Natural Language Process N-gram Model application, the model designs a compact disk index struc-ture Bc base on memory operations to make up the inadequate of traditional B tree.JBc-LSM achieves good performance on database building and updating, however its online indexing performance is common.According to the characteristics of JBc-LSM and the SSD, Flash-JBc-LSM system is proposed. This topic first introduces two main indexing system structure, then introduces and compares the main implement methods of LSM model. Combining the analysis of JBc-LSM system, then this topic introduces SSD to the system as the cache level to speed up the performance of the index system. In this paper, we extend a three-layout buffer on the basic of JBc-LSM, the first buffer layer is memory index structure Judy, and the second buffer layer is record structure on SSD, the query of record through the memory hash table indexing. The third buffer layer takes advantage of the Facebook Flashcache base on the Linux Device Mapper technology, this layer will cache the disk data in SSD in block unit.Comparing with the several mainstream schemes, Flash-JBc-LSM system proposed by this topic has better performance than other systems from the experimental results. And the experimental results also prove that the introduce of SSD and the indexing structure for SSD contribute to the system performance and NLP N-gram Model applications. This paper also discusses the future of the research topic base on the experimental results.