You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello @ib31iat, thanks for bringing this to our attention.
It seems to me that you are correct - Carbontracker in its current state only reads package RAPL files /sys/class/powercap/intel-rapl:0/energy_uj whereas DRAM is located in a subdirectory to this (for example /sys/class/powercap/intel-rapl:0/intel-rapl:0:1/energy_uj).
The RAPL zone hierarchy, clarified in this issue on the powercap repository, specifies that the energy_uj for the package zone does not include DRAM even though it is a subdirectory to the package directory. The original authors of carbontracker believed otherwise at the time it seems. Luckily, DRAM does not contribute much to the total energy consumption of a system.
I have made a fix for this - now it should measure both the package and DRAM files. Am testing on different devices now to validate.
Do I see correctly that CarbonTracker only reads the the
package
RAPL file (/sys/class/powercap/intel-rapl:0/energy_uj
) if it is available? On most machines this does not include DRAM energy consumption (see: https://hubblo-org.github.io/scaphandre-documentation/explanations/rapl-domains.html).Would that mean that CarbonTracker, contrary to what is written in the paper, only consider the Power of the CPU or do I miss something?
The text was updated successfully, but these errors were encountered: