Minimizing read bottlenecks in I/O bound systems, Effective and adaptive I/O using C++

Abstract

The overall performance of a computer system that reads a lot of data is directly linked to its ability to extract performance from the storage hardware. This thesis, conducted together with the company Carmenta, examines how to optimize a system’s data reads. The targeted storage device is an NVMe SSD. Several experiments simulating different workloads were conducted and laid the foundation for a new set of guidelines. Lastly, the guidelines were evaluated in a custom-built benchmark. Based on the evaluation, it was concluded that by using the guidelines, a system’s performance could be increased.