cpu_stats

python files for efficiently capturing multi-core cpu usage

we use the psutil module (documentation: https://psutil.readthedocs.io/en/latest/)

step 1) install dependencies (psutil)

pip install -r requirements.txt

step 2) record cpu stats, sampling cores every 10ms

python3 cpu_stat.py --sample 0.01 --binary_output mystats.bin

step 3) convert the binary file to a csv (with header)

python3 cpu_stat_to_csv.py --binary_input mystats.bin --header >mystats.csv

step 4) open the jupyter notebook and make sure to load the csv file by changing the following line

dataframe = pandas.read_csv("mystats.csv")

NOTES

1. you can monitor a specific process using the --pid parameter; however this does not break down the usage of each CPU rather the overall usage

2. estimated overhead for cpu_stat.py (measured on a 12 core osx)

   Sample rate (seconds)	CPU % / num cores	CPU %
   0.01	0.3	4
   0.1	0.035	0.45
   1	0.01	0.1

TODO

1. record user/kernel(sys) splits
2. spawn process and monitor pid