How to properly trace system characteristics #221
Description
I wanted to ask for your experience and suggestions regarding tracing the RAM and CPU usage of the program.
I know that the dkms modules for kernel tracing exists in LTTng and is supported by ROS 2, but when I try to use it, I only get millions of lost events according to TraceCompass, which make it impossible to use. This is also true for the user space libc tracing...
I am working on a constrained system (KRIA KR260 FPGA) so I would like to get some rough information regarding if certain optimizations/settings reduce CPU or RAM usage.
Do you have any alternative ideas? Is it perhaps possible to trace changes on /proc/meminfo or /proc/stat perhaps even a bit less often such that it doesnt affect system performance?
My current best idea is to add diagnosis ROS 2 messages, record the bagfile and merge it with the traces later but this doesnt sound very efficient..