A Quantitative Analysis on Semantic Relations of Data Blocks in Storage Systems

Abstract

In a big data era, there are and will be more demands and higher requirements for data organization and management in the storage system. Previous studies on semantic relations of blocks have set an essential theoretical basis in storage systems. Revealing semantic relationship of blocks has great scientific values and indicates a promising prospect in the application. However, only qualitative description of block correlations has been obtained by traditional ways, and few researches have been done on unifying and quantizing the blocks semantic relationships. In this paper, an uniform design-aided gene expression programming (UGEP) method has been proposed to address this issue. After analyzing historical access streams, a quantitative description system on semantic relationship of blocks is established. The semantic relationship of blocks can also be used in prefetching in storage system, which means the next access block can be calculated by the prefetching model. Simulation tests in four real system traces indicate that compared with sequential prefetching and no prefetching model, the quantitative semantic relationships model based on UGEP can highly improve the performance of the storage system. The experimental results show that in Cello-92 cases, compared with no prefetching method and sequential prefetching method, prefetching based on UGEP can reduce average I/O response time by 23.7% and 17.9%, and hit ratio is also increased by 16% and 8% as well.