Skip to content
This repository has been archived by the owner on Nov 11, 2022. It is now read-only.

Reduce gossip on past rounds #182

Closed
tomusdrw opened this issue May 13, 2021 · 3 comments · Fixed by paritytech/substrate#10882
Closed

Reduce gossip on past rounds #182

tomusdrw opened this issue May 13, 2021 · 3 comments · Fixed by paritytech/substrate#10882
Labels
enhancement New feature or request M-gadget The client code of the gadget.

Comments

@tomusdrw
Copy link
Contributor

tomusdrw commented May 13, 2021

Let's try to optimize the gossip a bit by sending votes for last round to all peers that we know (like right now),
but let's limit the number of peers that we send votes from rounds [last-5; last-1] to sqrt(peers).

This should reduce excessive gossip, but without being super complicated to implement.

@tomusdrw tomusdrw added enhancement New feature or request M-gadget The client code of the gadget. labels May 13, 2021
@ordian
Copy link
Member

ordian commented Jul 2, 2021

@andresilva I'm not familiar with beefy, would it make sense to adapt an approach similar to paritytech/substrate#9131?

@adoerr
Copy link
Contributor

adoerr commented Jul 5, 2021

@ordian it makes totally sense to adapt the GRANDPA gossip limiting / reducing strategies for BEEFY as well. We are planing to do that gradually over time.

Currently, we can not reduce gossip too aggressively because we first have to hook into the block import queue in order to be able to ask for / process missing justifications.

@tomusdrw tomusdrw added this to the Get BEEFY in shape milestone Nov 26, 2021
@tomusdrw
Copy link
Contributor Author

The gossip should be completely overhauled according to https://github.com/paritytech/grandpa-bridge-gadget/blob/td-docs/docs/beefy.md#gossip

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
enhancement New feature or request M-gadget The client code of the gadget.
Projects
None yet
Development

Successfully merging a pull request may close this issue.

3 participants