Measuring non-idle time spent in process

General Usage
, , ,
1

Hi all!
I want to measure the non-idle time spent in a given process. If I use Julia’s wall clock with time(), other processes that run on my computer (like parallel threads) will change the measurement. Judging by the code sample below, @timed seems to work, but I have not found another source. Can someone confirm this?

julia> using Base.Threads

julia> nthreads()
10

julia> function f()  # dumb function
           t = @elapsed for i in 1:100000
               exp(rand())
           end
           return t
       end;

julia> function g(N)  # single-threaded loop
           T = Vector{Float64}(undef, N)
           for n in 1:N
               T[n] = f()
           end
           return T
       end;

julia> function h(N)  # multi-threaded loop
           T = Vector{Float64}(undef, N)
           @threads for n in 1:N
               T[n] = f()
           end
           return T
       end;

julia> @timed g(1); @timed h(1);  # compile once

julia> res_g = @timed g(nthreads());

julia> res_h = @timed h(nthreads());

julia> total_time_g = res_g.time;

julia> total_time_h = res_h.time;

julia> total_threads_time_g = sum(res_g.value);

julia> total_threads_time_h = sum(res_h.value);

julia> total_threads_time_g / total_time_g  # around 1
0.9975217449895315

julia> total_threads_time_h / total_time_h  # greater than 1
5.692441324049729
2

I’m also wondering what this package does exactly, and whether it fits my needs: GitHub - schmrlng/CPUTime.jl: Julia module for CPU timing

EDIT: apparently that is exactly what I needed.