Actix-based Erlay simulator

This is an Actix-based simulator for evaluation of Erlay transaction propagation technique.

It includes:

• Peer actor that can be either public or private
• RecSet -- a set with reconciliation using minisketch-rs
• TrafficCounter actor that reads reports on used traffic from peers
• Various protocol messages for connection, tx propagation and set reconciliation

Note that set difference estimation and sketch bisection are omitted in this simulator. However, this can change in future.

It creates a network of public and private peers. Each private peer announces a single transaction of 1024 bytes to a single random outbound peer.

Short IDs of the transactions are based on SipHash2-4 and are 64-bit wide.

Example

Let's create a simple network and simulate both Erlay and flooding transaction propagation strategies.

Flooding

Run simulator with following parameters:

cargo run -- --numprivate=8 --numpublic=2

It will produce following network of peers:

• Two (2) public peers that do have inbound connections
• Eight (8) private peers that don't have inbound connections

Simple transaction flooding will produce 429488 bytes of traffic.

Traffic per peer:

pub0: 73280 ↑ 23816 ↓ (bytes)
pub1: 67088 ↑ 67160 ↓ (bytes)
priv0: 9288 ↑ 15480 ↓ (bytes)
priv1: 9288 ↑ 15480 ↓ (bytes)
priv2: 9288 ↑ 15480 ↓ (bytes)
priv3: 9288 ↑ 15480 ↓ (bytes)
priv4: 9288 ↑ 15480 ↓ (bytes)
priv5: 9288 ↑ 15480 ↓ (bytes)
priv6: 9288 ↑ 15480 ↓ (bytes)
priv7: 9288 ↑ 15480 ↓ (bytes)

Erlay (low-fanout flooding + set reconciliation)

Now we enable Erlay which is using set reconciliation (-r flag) instead of flooding for tx propagation:

cargo run -- --numprivate=8 --numpublic=2 -r

In this case, overall traffic will be 357488 bytes.

Traffic per peer:

pub0: 75928 ↑ 21752 ↓ (bytes)
pub1: 75928 ↑ 23816 ↓ (bytes)
priv0: 3352 ↑ 16656 ↓ (bytes)
priv1: 3352 ↑ 16656 ↓ (bytes)
priv2: 3352 ↑ 16656 ↓ (bytes)
priv3: 3352 ↑ 16656 ↓ (bytes)
priv4: 3352 ↑ 16656 ↓ (bytes)
priv5: 3352 ↑ 16656 ↓ (bytes)
priv6: 3352 ↑ 16656 ↓ (bytes)
priv7: 3352 ↑ 16656 ↓ (bytes)

As we can see, Erlay benefits us with 100 - 357488.0 / 429488.0 * 100 = 16.8% bandwidth reduction, mostly for private nodes.

Graphs

1. Used bandwidth in bytes depending on number of private nodes (1-500) with two (2) public nodes:
2. Used bandwidth in bytes depending on number of private nodes (10-300) with eight (8) public nodes: