Skip to content

Conversation

@frankmcsherry
Copy link
Member

This PR introduces a broadcast channel allocation method to communication::Allocate. It defaults to using a point-to-point channel and cloning the contents, but it can be implemented more thoughtfully by implementors. For example, we might soon like to have the inter-process channels only send the data once, rather than once for each remote worker.

Potential follow-up: an all_reduce channel, not unlike a broadcast channel but with the opportunity to consolidate updates that move along the channel. Both the broadcast and hypothetical all-reduce are in service of progress messages, which support an all-reduce pattern.

@frankmcsherry frankmcsherry merged commit 42fe8bc into TimelyDataflow:master Feb 4, 2025
7 checks passed
@github-actions github-actions bot mentioned this pull request Feb 4, 2025
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

Successfully merging this pull request may close these issues.

1 participant