10
votes

I am confused by goroutine, user thread, and kernel thread concepts

  1. From effective go introduce goroutine, so what does os threads means which mentioned by the paper? Does it mean user thread or kernel thread?

  2. From go-scheduler paper, I learn about M G P, and why the numbers of P is equal to the numbers of CPU? If all the cpus serve for the go program, but other program in the os system have no cpu thread to execute?

  3. How many kernel thread generated by the os system?

1
I think you should take measures to gain more generic knowledge about how a contemporary commodity OS shares CPU time and other resources between the processes running on it. You can start here but a good introductory-level book should also help.kostix

1 Answers

14
votes

Let's include the picture from the go-scheduler page you linked to.

enter image description here

And establish the terminology:

  • M: OS thread, can also be called a kernel thread
  • P: processor, or scheduling context
  • G: Goroutine

goroutines are what we are most familiar with in Go, and could be considered user threads. A more technical name for those is Green Threads.

P is used to perform the mapping from many goroutines to many OS threads. There is one per OS thread, and the number is determined by the value of GOMAXPROCS (by default, the number of CPUs as reported by your system).

So, to answer your questions in order:

  • OS thread means kernel thread
  • GOMAXPROCS defaults to the number of cores, but you can change that. Just because you can run on all cores does not mean you're not leaving CPU time for other processes. Concurrency usually involves a lot of waiting for IO. Even if you're going crazy hashing things, the kernel scheduler will boot you off to run other things.
  • there are as many OS threads as needed. Looking at ps -eL, my system currently has 1434, some of them actual kernel jobs, some for my go program.

You can find a really good explanation of OS vs Green threads in this answer