I understand that optimal swappiness depends on the application. Database servers often have a swappiness close to zero, while in many other cases it is apparently recommended to keep the default value of 60. However, I am not sure which setting suits my scenario.
The server runs Ubuntu 20.04, has 8 CPUs, around 30 GB RAM, and an SSD drive. I use it to execute parallelized (SOCK cluster) R scripts usually involving geo-spatial data. While running the R script, I do not execute any other application on the machine. And I read the data directly from a file on disk, not from any SQL or other database system. With the current task, the process exceeds the available RAM every few minutes for a few seconds and otherwise remains well below that limit. There are likely some inefficiencies in the package I am using. However, I will neither adjust the package code, nor will I install further RAM. What I would like to do is to set a swappiness value that best suits my application.
A rule of thumb regarding such data science applications would be nice.
