Network Infrastructure

4. Network Infrastructure

4.1 Virtual gateway

To keep the consensus layer stable in an open network, Seth puts a virtual gateway at the entry point. Before traffic reaches shard consensus and execution, it goes through a simple pipeline: quick checks → quota shaping → protocol routing. The gateway first filters transactions and messages early (for example, signature validity, basic format and size checks, and simple balance/nonce rules) and drops clearly invalid or malicious load at the edge, so it does not consume consensus or execution resources.

It then rate-limits and shapes traffic by connection, address, and subnet, compressing bursts into a controllable range and reducing the impact of DDoS and spam on liveness.

Finally, it routes transactions and finalized receipts to the correct shard and pool, so inbound load is structured before it hits execution, lowering the chance of congestion and hotspot breakthrough.

4.2 Random layered broadcasting

In large node networks, plain gossip creates many duplicate forwards. This wastes bandwidth and can overload weak connections. Seth uses Random Layered Broadcasting (RLB): in each round, a node forwards a message to only 𝐾 randomly chosen neighbors, and uses deduplication and simple path constraints to avoid loops.

This changes broadcast behavior from “flood all neighbors” to “layered diffusion.” Under current test constraints (fixed 𝐾 with dedup enabled), redundancy stays around 1:8, which reduces mobile data use and improves propagation stability. A comparison of gossip and RLB is shown in Figure 2.

_{Figure 2. Comparison — Gossip vs Random Layered Broadcasting}

4.3 Bitmap height tree

To reduce catch-up cost when mobile nodes disconnect and rejoin, Seth uses a bitmap height tree to speed up missing-block localization. A node represents block-height presence as a bitmap, compresses it by fixed-size segments (for example, every 64 heights into one uint64), and then builds a root-level summary over the segment digests. During synchronization, nodes first exchange these summaries, then use bit operations to quickly locate missing height ranges and fetch only what is needed.

This avoids repeated hashing and proof traversal overhead typical of Merkle-based approaches. The result is near O(1) work for localization (bit operations plus a small number of summary exchanges), which fits weak networks and low-power devices better.

_{Figure 3. Comparison — Gossip vs Random Layered Broadcasting}