This repository contains the source code associated with our paper "PeerSwap: A Peer-Sampler with Randomness Guarantees", accepted at SRDS'24. PeerSwap is a peer-sampling protocol with provable randomness guarantees.
First install the required Python dependencies:
pip install -r requirements.txt
The simulator can be executed in two modes.
In basic mode, we first generate a k-regular graph using networkx, and then schedule peer swaps according to the Poisson clock on each edge.
While executing the swaps, we keep track of the neighbourhood distributions of a single, random peer or all peers if the --track-all-nodes option is provided when starting the program.
Basic mode can easily be enabled by passing the --basic flag but does not support network delays or clock desynchronization between peers since it assumes swaps are executed instantly.
For example, you can use the following command.
python3 main.py --cpus 1 --runs-per-process 100 --nodes 1024 --time-per-run 7 --k 7 --basic
The above command will run 100 experiments sequentially and each experiment will run PeerSwap for 7 seconds using a 7-regular graph with 1024 nodes. Your output should look something like this:
Will start experiments on 1 CPUs...
Experiment took 7.301254 s., swaps done: 2509524
Processes done - combining results
At the end of the experiment we combine the individual results of all runs.
If you don't pass the --cpus option, PeerSwap will use all available CPUs minus two on the system.
Default argument values can be found in peerswap/args.py.
By default however, we execute the full PeerSwap protocol where peers exchange a sequence of messages to perform a single swap. This mode is much slower than basic mode but it enables the evaluation of PeerSwap in the presence of network delays. For example, see the following command.
python3 main.py --cpus 1 --poisson-rate 0.1 --max-network-latency 0 --runs-per-process 100
Your output should look something like this:
Will start experiments on 1 CPUs...
Experiment took 6.155684 s., swaps done: 24603, failed swaps: 0
Processes done - combining results
In default mode, PeerSwap will generate pairwise latencies uniformly between 0 ms and a maximum latency specified by the --max-network-latency option (in seconds).
One can also specify a latency matrix as a CSV file with the --latencies-file option.
For example, to specify the latencies for a three-node network, your CSV file can look like as follows (note that the latencies are provided in milliseconds):
0,183.6055,90.8835
183.029,0,44.826
90.7965,44.793,0
If there are more nodes in the experiment than rows/values in the latency matrix, PeerSwap will automatically assign nodes to latency values in a round-robin fashion.
If you use our work, please cite us using the following citation:
@inproceedings{guerraoui2024peerswap,
title={PeerSwap: A Peer-Sampler with Randomness Guarantees},
author={Guerraoui, Rachid and Kermarrec, Anne-Marie and Kucherenko, Anastasiia and Pinot, Rafael and de Vos, Martijn},
booktitle={Proceedings of the 43rd International Symposium on Reliable Distributed Systems (SRDS 2024)},
year={2024}
}