Parallel broadcast on unreliable medium

Question

I have a hundreds or thousands of nodes, who all know each others addresses.

In order to efficiently spread a message between them, I'm currently thinking that I need to turn them into a tree, or perhaps multiple trees.

I've read https://en.wikipedia.org/wiki/Broadcast_(parallel_pattern) but it seems to not handle message loss.

Is there a way to deliver to 99% of nodes if messages have a 1% chance of failure? Speed is important here, so retries and/or TCP are out.

A simple tree broadcast will lose whole branches when a message fails. Perhaps multiple trees would solve the problem? How many?

I am happy for nodes to receive a small number of redundant messages. I need to use UDP. A small number of absolute failures is OK, as they can catch themselves up at a later time, dropping off of the network in the meantime.

Is this a solved problem?

(I've come across https://ieeexplore.ieee.org/abstract/document/508133 but can't afford to pay before knowing whether it's suitable.)

Answer 1

Here are some interesting methods in distributed system like gossip, this can solve your problem.