Review Club: P2P - Increase tx relay rate

[i-am-yuvi]

Session Details:

Date: 10-10-2024
Time: IST 20:00 (UTC 14:30)
Link: bitcoin/bitcoin#28592
Difficulty: Easy

Motivation:

The Bitcoin network traditionally operates with an expected transaction relay rate of approximately 7 transactions per second. However, this baseline assumption only partially accounts for network behaviour when processing smaller transactions, which can be relayed more efficiently. When the network processes smaller transactions, it can actually sustain higher relay rates than the standard 7 tx/second. This capability, while beneficial for network throughput, creates an unintended consequence: the per-peer inventory queues can become excessively large, potentially leading to memory management issues and reduced network efficiency.

Notes:

Inbound Peer Rate

• Previous rate: 7 tx/second
• New rate: 14 tx/second
• Factor of increase: 2x
• Rationale: Provides improved safety margin for handling increased transaction volumes

Outbound Peer Rate

• Maintains existing 2.5x multiplier relative to the inbound rate
• Previous rate: ~17.5 tx/second (7 * 2.5)
• New rate: 35 tx/second (14 * 2.5)
• Rationale: Preserves established outbound prioritization while scaling with the new baseline

Questions:

• What would be the effect of this change in the network?
• What if the Tx relay rate was too high?
• What if the Tx relay rate was too low?
• Which simulation tools can be used to test this change?

Learnings:

p2p
transactions
tx relay rate

[i-am-yuvi]

Summary:

For each transaction that a user submits to the network, asymmetric bandwidth usage is required for the transaction to propagate to the entire network. Without mitigations, network bandwidth could be intentionally abused or accidentally overused.

The minimum fee rate helps mitigate these problems by preventing “free relay” attacks and ensuring that bandwidth usage has a minimum cost in satoshis. The tx relay rate is also important because it essentially limits the multiplier of each node on incoming traffic to outgoing traffic. Currently, every transaction we see will INV every other node (which would change if we adopt erlay).

One one side: if the tx relay rate didn’t exist or was too high, an attacker can pay the minimum fee rate to use greater than [num nodes on network] x [size of txn] amount of bandwidth on the whole network.

On the other side: if the tx relay is too low, transaction propagation is laggy, and there will be dramatic differences in the mempools of different nodes. The rate being “too low” is in relationship to the (tx creation - block creation) rate, aka how many txs are hanging out in the mempool over time.

Too High Relay Rate

Risks:

• Bandwidth amplification attacks possible
• Network congestion
• Resource exhaustion
• Disproportionate cost/impact ratio

Too Low Relay Rate

Problems:

• Slow transaction propagation
• Mempool inconsistency
• Fee market inefficiency

Also,
if transactions are being created with a random fee rate distribution, that is probably a bit different than what we naturally see out in the wild. on mainnet we usually see tx fee rate rising as the mempool gets more full. This is relevant because we’ve identified synchronization of the top of the mempool as a key metric.

TXs with random fee rates will organically lead to a stronger synchronization of top-of-mempool in a short amount of time while increasing fee rates will see stronger divergences in short-time intervals.