The core architecture is split into two components: a timestamping signing service and a P2P gossip network. By decoupling the gossip network, I'm simulating its performance using a Monte Carlo approach. With a basic gossip protocol, the simulation reaches ~10k TPS on a 100-node, randomly connected network (not fully connected), and I see a lot of potential for further protocol optimizations.
Initially, I considered a more resource-intensive approach using Shadow [2] for more realistic node simulations, as outlined in this discussion on libp2p's Gossipsub stress metrics [3]. However, the Monte Carlo method allows me to simulate the network more efficiently without needing to deploy full nodes.
In parallel, I'm exploring game-theoretical concepts for selecting signers and ensuring the system remains open to new entrants. One paper I'm currently diving into is "Collusion, Efficiency, and Dominant Strategies" [4].
[1] https://news.ycombinator.com/item?id=41687715
[3] https://discuss.libp2p.io/t/rough-stress-metrics-for-gossips...
[4] https://www.sciencedirect.com/science/article/pii/S089982561...