how to analyze cpu time while benchmarking in python (multiprocessing)?

Question

Followings are the result after profiling using %time in ipython-

single-thread: CPU time: user 6m44s sys 1.78s total 6m46s Wall time: 5m19s

4-thread: CPU time: user 10m12s sys 2.83s total 10m15s Wall time: 4m14s

Shouldn't CPU time be lesser for multi-threaded code ? Also, how can be CPU time be more than wall time, as wall time is total elapsed time. Could you please clarify these terminology.

Answer 1

No CPU time is the time spent by all cpus on the task. So if cpu1 spent 2 minutes on a task and cpu2 spent 3 minutes on the same task the cpu time will be 1 + 3 = 4.

So in multithreaded programs we would expect that cpu time will usually be more than the wall time.

Now you might ask why does the same hold for your single-threaded program. The answer will probably be that even if your code does not explicitly use parallelism, there is probably a library you use that does.