Improve your storage I/O performance today

an article by Lucas Crämer and Jannik Smidt

DISCLAIMER

This post tries to keep the complexity manageable while making a point clear. We are not systems engineers/kernel developers, so feel free to point out any mistakes/misunderstandings. This post probably does not present anything new for people who are working in systems engineering and/or kernel development.

Why is that of any interest to an application developer?

Storage is getting faster and faster, while the performance of single CPU cores does not increase to the same degree. Blazingly fast NVMe Storage Devices have reached PCs, Smartphones, Gaming Consoles, and Server Systems in the Cloud. The importance of fast and low latency storage in the cloud can not be understated. AWS recently even went that far to develop and release their own SSDs (AWS Nitro SSD) based on the AWS Nitro System for specific EC2 instance types [1]. 

Many application developers are not particularly educated about storage hardware, kernels, and file system APIs. This post is basically what application programmers should know when programming with modern storage devices in mind.

Continue reading