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

[Feature Request] Network Statistics #520

Open
chavinlo opened this issue Nov 18, 2022 · 4 comments
Open

[Feature Request] Network Statistics #520

chavinlo opened this issue Nov 18, 2022 · 4 comments
Labels
enhancement New feature or request help wanted Extra attention is needed

Comments

@chavinlo
Copy link

Is your feature request related to a problem? Please describe.
No.

Describe the solution you'd like
Basically display 2 stats:
1.- Network Input/Output in mbps
2.- Live count of gradient % sync

Describe alternatives you've considered
At the moment I plan on just using an external tracker but it would be nice to have this embedded

Additional context
I'm using pytorch lightning as I had an issue with the native version of hivemind (#519 )

@chavinlo chavinlo added enhancement New feature or request help wanted Extra attention is needed labels Nov 18, 2022
@borzunov
Copy link
Member

borzunov commented Nov 18, 2022

Hi @chavinlo!

Thanks for the feature request! We're busy with another project built on top of hivemind at the moment, but will consider adding this feature when we get back to working on the hivemind core.

Currently, we just use external utilities like nvtop to monitor the traffic (input/output in Mbps). It doesn't display the progress of the gradient sync directly, but one usually can get a sense of it by watching the spikes in traffic (e.g., if previous gradient syncs took 30 sec and the current spike lasts for 15 sec, you know that the sync progress is 50%).

I agree though that an explicit way to watch the progress would be much more convenient.

@ikmckenz
Copy link
Contributor

ikmckenz commented Dec 9, 2022

Would the desire be to monitor at the p2pd layer, or at the Python layer?

libp2p has some metrics built in: https://github.com/libp2p/go-libp2p/tree/master/core/metrics

But it might be easier to just do it inside the Python DHT class?

@Bandcompute01
Copy link

We are building this feature into our platform and can send it to you when it is ready?

@mryab
Copy link
Member

mryab commented Dec 13, 2023

Hi @Bandcompute01, sure, that would be awesome! If you are building this on top of hivemind, it would even better to integrate this into the repository with a pull request: if you're interested, I am happy to assist with that

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request help wanted Extra attention is needed
Projects
None yet
Development

No branches or pull requests

5 participants