Goals

[jannikluhn]

Quantities of interest:

• Number of hops required for messages to reach >x% of all subscribed nodes
• Peak and average bandwidth load for validators and shard nodes
• Impact of bandwidth and latency limits on propagation times
• Reliability of the protocol if some fraction of nodes (either of full network or of a single shard) goes offline
• Reliability if node number of a single shard is low (to what extent do validators contribute to keeping shard alive?)
• Time it takes to connect to new shard (valid question at least for gossip sub, other protocols might not specify discovery)

Compare different protocols considering above metrics:

• FloodSub
• GossipSub (difference might not matter if we're limited by latency instead of bandwidth)
• Combinations of above (for different message types)
• Alternatives?

Explore different topologies/connection strategies:

• One beacon chain network/topic and one network/topic per shard
• Only shard networks (beacon chain piggy-backs on shard networks)
• Validators connect by default to random "home shard"
• Nodes connect by default to multiple shards

[jamesray1]

While I'm working with Rust on gossipsub, I'm interested in your findings at a design level!