chakib-belgaid/net-energy

Net energy consumption of program

a script that will use powerapi/hwpc-sensor to gather the average power into the machine , and substruct it from the programs global energy conusmption

Steps

• install perf

  sudo apt install linux-tools-common gawk

• install the required python libraries

• Launch hwpc-sensor with the watcher.sh

-use the script measure.sh to measure the enregy of your work

-./measure.sh something

Remark

To avoid lack of space in the computer we create a cleaner that will delete the reporting file each day

1. create the cleaner script and put it in /opt/scripts/watcher.cleaner.sh : the script will contain this instruction

   echo "" > /tmp/powerapi-sensor-reporting/watcher/rapl

2. add it the a schedular using the command crontab -e for this u need just to add his absolute path to the config file

Workers

In order to simulate real production environment we create a script that customise the worload of the cpu

it accepts two parameters

• the workload of the cpu , a value from 0-100 : 100 means 100% of the ressource

• the number of threads to be used

  python workers.py workload threads_counts

Demonstration

the reasult of the script meansure.sh would be like the following image

• av-system-power (Watts): the average power consumption of the machine before we run our program
• av-program-power (Watts): the average power consumption of the machine during the execution of our program
• system-energy (Joules): the estimated energy of the machine wihtout our program (we calculate it with v-system-power * execution-time )
• program-energy (Joules): the energy consumption of the machine including our program
• program-net-energy (Joules): the net energy that our program consumed during his execution
• energy-cores | energy-gpu | energy-pkg (Joules): the raw measures gathered from perf -this will help to estimate the error of the program-energy since it is just an estimation
• execution-time (seconds): execution time of our program