Gossip Protocol Simulation in NS-3

Overview

This project implements a distributed gossip protocol for signature distribution in a hierarchical network using the NS-3 network simulation framework. The simulation is parallelized using MPI (Message Passing Interface) to distribute computations across multiple processes, allowing for efficient simulation of large-scale networks.

System Architecture

The system is organized as a hierarchical network with three levels:

1. Peers: Basic entities that create and gossip signatures.
2. Subnet Aggregators: Collect signatures from peers within their subnet and forward them to the global aggregator.
3. Global Aggregator: Receives aggregations from all subnet aggregators.

Gossip Protocol

Instead of peers sending signatures directly to subnet aggregators, signatures are gossiped between peers with a fan-out factor of 6. Each peer:

• Generates its own signature
• Forwards new signatures to 6 randomly selected peers in the same subnet
• Only forwards each unique signature once

Subnet aggregators monitor the signatures spreading through the network and collect them until reaching a threshold (2/3 + 1 of all peers in the subnet).

MPI Parallelization

The simulation leverages MPI to distribute the computational workload:

• Each MPI process handles a subset of subnets
• Only rank 0 (master process) hosts the global aggregator
• Workers communicate with the master through MPI messages
• Each process simulates only its assigned subnets, reducing memory usage and improving performance

Requirements

• C++23 compiler
• CMake 3.25 or higher
• MPI implementation (OpenMPI recommended)

Setup and Installation

Step 1: Download and Build NS-3

The project includes a script to download, build, and install NS-3.44 locally within the project:

# Make the setup script executable (if not already)
chmod +x setup_ns3.sh

# Run the setup script
./setup_ns3.sh

This script will:

1. Create an external directory
2. Download NS-3.44 source code
3. Extract and build NS-3 with optimized settings
4. Install NS-3 within the project folder

The process may take 20-30 minutes depending on your system.

Step 2: Build the Project

mkdir -p cmake-build-debug
cd cmake-build-debug
cmake ..
make

Running

To run the simulation with MPI:

mpirun -np <num_processes> ./main [options]

Replace <num_processes> with the number of MPI processes you want to use.

Command Line Parameters

You can configure the simulation using the following command-line options:

Parameter	Description	Default
--nSubnets=<value>	Number of subnets in the simulation	5
--nPeersPerSubnet=<value>	Number of peers in each subnet	512
--nRounds=<value>	Number of rounds of aggregation to run sequentially	5

Examples

# Run with 4 MPI processes using default settings
mpirun -np 4 ./main

# Run with 8 MPI processes, 5 subnets, 512 peers per subnet
mpirun -np 8 ./main --nSubnets=5 --nPeersPerSubnet=512

# Run with 16 MPI processes, 16 subnets, 1024 peers per subnet, 3 rounds
mpirun -np 16 ./main --nSubnets=16 --nPeersPerSubnet=1024 --nRounds=3

Configuration Parameters

The main configuration parameters are in main.cpp:

• nSubnets: Total number of subnets in the simulation
• nPeersPerSubnet: Number of peers in each subnet
• m_fanOut: Number of peers that each peer gossips to (default: 6)
• m_threshold: Threshold for subnet aggregators (default: 2/3 * nPeers + 1)

Performance Metrics

The simulation outputs several performance metrics:

• Total virtual time: Time taken in the simulated environment
• Total real execution time: Actual wall-clock time
• Ratio (virtual/real): Efficiency of simulation
• Total number of peers: Size of the simulated network

Logs Format

All logs include timestamps in milliseconds and identify the source:

• Peer messages show the peer ID
• Subnet Aggregator (SA) messages show the subnet ID
• Global Aggregator messages show when subnet aggregations are received