Perhaps it would be even better to mention what the number after '-j' means.

Very good point. I added a short description (as I don't understand the gory details myself) and also added a link to the official GNU make parallelism documentation for those interested in the details. Do you think that's excessive?

I think we should add -j4 directly to the example above (the ./configure then make).

This is based on what I saw at the code and learn; sure it would be great if people read more but typically they start running stuff in order (which is reasonable for people who are new to something).

So... it seems somewhat odd to tell people to build without it and then come back around and say to do it with it. It would be best for the general experience for it to be directly there, with notes if it does not work and for additional processor cores.

I guess the question is whether using make -j4 (or even -j8) ever causes appreciable slowdown or other issues on the kind of machine that people who are new to make are using. I'd argue that it won't, and it will help the vast majority of people. If you're running on a machine that can't handle 4/8 threads, then you probably know what you're doing with make.

If most people have 2/4 core hyperthreaded laptops, I'd say just use make -j8 like we do in CONTRIBUTING.md.

See also #8286.

There is some discussion there but from my observations the average appears to be 4 threads (some have less or more).

-j8 is nice if you can do it but it may be a little hard on lower-end laptops?

I have a coworker who just got a new dual core MacBook Pro (I believe it is hyper-threaded). I'll ask him to run make with -j8, -j4, -j2, and without parallelism today and get some metrics on how long each took.

I realize that it will be extremely unscientific, but it may give us a bit of insight.

Side note, it would be really cool if we could figure out a way to gather some metrics on what kind of machines people are using to develop node and node projects. Maybe eventually we can send out a hardware survey like the user survey that was just sent out?

This actually came up during the code and learn for someone with a MacBook Air. We had to switch to -j4 since they couldn't even use their browser with -j8

Really Really Unscientific Benchmark Results:

Values taken from running $ time make -j[n]. All builds were run immediately after deleting the ./out directory. Both computers were 100% usable during the build without any signs of slowdown and were being used for web browsing and simple text editing.

I ran out of time with my coworker's laptop to run it without concurrency but that's kind of irrelivent to the current conversation anyway.

Mid 2012 Macbook Pro - 2.3GHz i7 - Hyper-Threaded Quad Core

make -j8 2746.33s user 260.92s system 601% cpu 8:19.88 total

make -j4 1805.77s user 192.59s system 370% cpu 8:59.70 total

make -j2 1721.45s user 165.89s system 191% cpu 16:24.13 total

Late 2016 Macbook Pro - 2.0GHz i5 - Hyper-Threaded Dual Core

make -j8 1904.60s user 150.79s system 347% cpu 9:51.45 total

make -j4 1851.63s user 145.75s system 339% cpu 9:48:07 total

make -j2 1244.67s user 115.51s system 183% cpu 12:20.78 total

Running with -j4 makes quite a bit of difference on both machines, but stepping up to -j8 doesn't really affect the dual core and only gives a minimal boost to the quad core.

Seeing these extremely unscientific results, and taking into account that -j8 is known to cause issues on at least one configuration, I think I'd personally lean toward recommending -j4 as it gives the best "bang for the buck" and is less likely to cause issues on lower end machines.

I went ahead and changed the docs to use -j4 for this PR as it seems like this is probably the best option. I'd be happy to change it back to -j8 if that's deemed to be the better option.

If we want to standardize on -j4 I'd be happy to go through the docs and update them to be consistent either in this PR or in a separate issue.

It also may be worth the effort for us to have our own doc on make that we can link to wherever the -j[n] flag is used in the documentation, though this may be a bit overkill.

If we want to standardize on -j4 I'd be happy to go through the docs and update them to be consistent either in this PR or in a separate issue.

+1, might as well do it in this PR

@bnoordhuis LGTY now? This looks land-able to me!

@jmdarling This isn't patching cleanly for me, do you think you'd be able to squash this into two commits (one for the note, one for the standardization of -j4)?

Sorry for the slow turnaround on that. My git-fu isn't what it should be and that simple task turned into an arduous journey 😝.

Landed in d8c7534...0cd1f54, thanks for the contribution! :D

the v4.x building.md does not include any information regarding parallel make. Please feel free to backport.