This paper discusses some ideas of how to extend HyParView/Plumtree beyond the most straightforward implementation. It presents the goals that we wanted to have in mind when building actual implementation in Rust.
While Partisan - the original and probably the most widely adopted implementation of HyParView+Plumtree - value proposition relies on building huge clusters (having >10 000 nodes) capable of broadcasting messages via self-healing broadcast trees, its network protocol seems to limit its possibilities to centralized solutions - like services living in data centres or at the edge, where peer discoverability and reachability is in control of their developers/operators. Other issue is that when exchanging neighbors/shuffle peers, plumtree blindly replaces them with current passive view without checking if they are reachable for the current peer in the first place.
We think, that adopting HyParView/Plumtree in Rust with extensible API could expand its capabil