future.callr::callr Behavior Under Windows 10 and Linux Debian

[drag05]

I have written a Shiny app that sends some SQL queries to a MySQL database and retrieves the results in tables. The queries can be sent and received in parallel. The Shiny app tabs are mutually independent: the User can drill down the results in one tab or send a query while other tabs are still processing queries.
Each tab is built as a Shiny module.
My current strategy is
plan(future.callr::callr, workers = parallelly::availableCores() %/% 3L)
and I use a 16-core CPU Windows machine with 64GB RAM

I also have Shiny Server installed on Windows Subsystem for Linux (in my case, Debian) virtual machine

What I am observing:

• Running the Shiny app on Windows (Rstudio)

with 3 parallel requests sent to the database , Windows Performance Manager shows between 6% and 15% CPU usage and 3 R processes running. The heat fans are spinning up

• Running the Shiny app on Linux VM (Shiny Server)

with 3 parallel requests sent to the database, Windows Performance Manager shows the vmmem service using between 1% and 3% CPU resources and several MB of memory. Heat fans are barely turning

The virtual machine has access to all 16 CPU cores:

$ cat /proc/cpuinfo | grep processor | wc -l

16

Questions:

• Do I use the correct plan for Linux?
• Should I use something else (like slurm) for MySQL database (which is located on cloud)?

Thank you!