Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Workers should adjust BLAS/LAPACK/Numpy threads #230

Open
ravwojdyla opened this issue Dec 29, 2020 · 6 comments
Open

Workers should adjust BLAS/LAPACK/Numpy threads #230

ravwojdyla opened this issue Dec 29, 2020 · 6 comments
Labels
question Further information is requested

Comments

@ravwojdyla
Copy link
Contributor

Should workers by default adjust BLAS/LAPACK/Numpy threads as per dask-best-practices? Since we have access to the number of threads/workers and cores, it should be possible adjust those accordingly. wdyt?

@jacobtomlinson
Copy link
Member

Absolutely. There is work happening already in dask/distributed#4377 to address this.

@ravwojdyla
Copy link
Contributor Author

@jacobtomlinson please correct me if I'm wrong, does dask/distributed#4377 really address this issue? Looking briefly into that, it seems like it's more about the number of Dask processes and thread but not the BLAS/LAPACK/Numpy threads?

@jacobtomlinson
Copy link
Member

Ah right I see, I misunderstood the request.

Most cluster managers support an env_vars configuration option, so folks can set things however they want.

@ravwojdyla
Copy link
Contributor Author

@jacobtomlinson thanks for prompt response. I understand, but do you think dask-cloudprovider could set sensible defaults given that it has information about number of Dask workers/threads and VM types (#CPUs)? wdyt?

@jacobtomlinson
Copy link
Member

That's is definitely a possibility. I'm not sure whether we do this in any other cluster managers. Perhaps someone else from @dask/maintenance might know?

@martindurant
Copy link
Member

Sounds like another point where spec-cluster might be able to give a consolidated view over different deployment types - like how many CPU cores are available to each worker, possibly different than the number of threads requested. I suspect, though, that in most cases setting all the threading options to 1 is the right thing to do.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
question Further information is requested
Projects
None yet
Development

No branches or pull requests

3 participants