JuliaPro is slower than the standard Julia. How come?

Tooling
1

After upgrading from JuliaPro 0.6 to JuliaPro 1.0.1 I noticed that the console starts slower, and loading a module for the first time in a fresh session is slower. At first I thought it was due to the new Julia version.

However, to try a package not yet included in JuliaPro’s registry but already included in the General registry, I installed the standard Julia 1.0.1 (the one on julialang.org not juliacomputing.com) yesterday. Since I have installed both I did a quick side-by-side comparison. Boy what a difference in loading time!

To start julia.exe in command line (no IDE involved):

  • JuliaPro: 10 seconds
  • Standard Julia: <1 second

“using PyPlot” in a fresh Julia session (already precompiled):

  • JuliaPro: 37.307783 seconds (41.05 M allocations: 2.514 GiB, 3.50% gc time)
  • Standard Julia: 8.733778 seconds (11.86 M allocations: 589.460 MiB, 4.22% gc time)

“using DataFrames” in a fresh Julia session (already precompiled):

  • JuliaPro: 15.906605 seconds (28.43 M allocations: 1.418 GiB, 6.10% gc time)
  • Standard Julia: 10.120652 seconds (19.96 M allocations: 1.028 GiB, 6.69% gc time)

Why such a big difference?

System information: JuliaPro 1.0.1, standard Julia 1.0.1, Windows 10 64-bit, Intel Core i7-6600U CPU with 8GB RAM.

3 Likes
2

I have the same issue…

3

Yeah, this is not right, we’re investigating. There absolutely should not be this large a difference.

Regards

Avik

4

Has this speed issue been addressed? I just installed JuliaPro 1.0.2 and I am new to the platform. The launch of JuliaPro was incredibly slow, and performance is not what I expected. Then again, I may be doing something wrong!

1 Like
5

No JuliaPro 1.0.2 unfortunately still has poor startup performance. JuliaPro 1.0.3 (which we release a couple days ago) is much better, but still not up to par. The next version will have fixed this completely.

To provide some additional context, this problem is caused by not precompiling some standard library modules such as Pkg (the package manager) correctly. i will note that this effects startup time, and the performance of the first pkg operations on the REPL. So while this undoubtedly results in a poor programmer experience, the code that you write should be no slower in JuliaPro compared to base julia.

Regards

Avik

3 Likes
6

Thank you, Avik.

7

is this already officially fixed?
I started using JuliaPro 1.2.0-1 however it feels sooo slow that I just started thinking about switching back to my own Juno.

8

Can you give some more info about what feels slow?

9

the interpreter takes minutes to react sometimes
especially slow:

  • adding new packages (inkl. scanning repository)
  • first shell startup

I now installed plain Juno in Atom with julia 1.2.0 and get similarly slow behaviour when firing the shell und for some evaluations/compilations. Also the adding new packages feels still slow, but I guess it is slightly faster than the JuliaPro variant.

A bit surprising that Juno has this slow parts.