search

rajveerb / lotus

Lotus: Characterization of Machine Learning Preprocessing Pipelines via Framework and Hardware Profiling
Other
3 stars 1 forks source link

Overhead/Accuracy comparison of our instrumentation vs other python profilers #25

Closed rajveerb closed 10 months ago

rajveerb commented 11 months ago

Compare overhead in terms of wall, sys, user time as well as storage overhead due to logs generated.

Moreover, we also compare the accuracy of the profilers.

We restrict the discussion to profilers which support multi processing tracking as our workload being profiled requires it, for instance data loader worker. Our target comparison is with below profilers:

  1. py-spy
  2. austin
  3. scalene
rajveerb commented 10 months ago

Currently, the log storage overhead has to be manually calculated.

This can be done by simply using du -b command to get file size for the log.