Readahead design is one of the crucial aspects of filesystem performance. In this paper, we analyze and identify the bottlenecks in the redesigned Linux 2.6 readahead code. Through various benchmarks we identify that 2.6 readahead design handles database workloads inefficiently. We discuss various improvements made to the 2.6 readahead design and their performance implications. These modifications resulted in impressive performance improvements ranging from 25% - 100% with various benchmarks. We also take a closer look at our modified 2.6 readahead algorithm and discuss current issues and future improvements.

...

Download PDF.