Improving Random Read Performance of Glibc

The Cloud data services, specifically, key/value stores and NoSQL database that require a large number of index lookups that fetch small amount of data. Random I/O becomes the critical performance factor. However, compared with sequential read, the efficiency of random read is very low. Our experiment will explain this. File I/O operation is closely associated with the implementation of I/O mechanism both in kernel space and user space. In this paper, we aims at analyzing the standard I/O mechanism, improving the standard I/O mechanism in user space for random read and implement into the glibc. Our experiment test result proves that our improved I/O mechanism will greatly improve the performance of random read.