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

Do we need Grpc for master and trainer communicate? #2364

Closed
dzhwinter opened this issue Jun 2, 2017 · 2 comments
Closed

Do we need Grpc for master and trainer communicate? #2364

dzhwinter opened this issue Jun 2, 2017 · 2 comments
Assignees

Comments

@dzhwinter
Copy link
Contributor

@helinwang metions that master needs to provide callback. As a result, he take Grpc into account for rpc library selecting.
#2363
here are my thoughts on master and trainer communicating process. It is not a whole design doc, just the part of communication.
In my mind, the trainer, master, pserver role can be written in go, and use go native rpc for communacting. The computation in paddle framework can be done in the optimizer library, which is written in c++. The topology defines part can be any glue language such as python, perl and so on.

@helinwang
Copy link
Contributor

helinwang commented Jun 2, 2017

This is a very good issue, we need more discussion in this area:

  1. Do we need the callback? Or maybe we can just use Tensorboard?
  2. What RPC to use.
  3. Does Jupiter notebook Python directly talk to the master, or Python talk to Go and GO talks the master.

I think it's a very good idea to bring this up. Maybe we can think about it while put most of the focus on our current work. Because this is still quite far (perhaps the last item we need to do for v1, and no other item is blocked by this one).

@helinwang
Copy link
Contributor

We are using go's net/rpc due to the simplicity.

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

No branches or pull requests

4 participants