Dandelion++ Simulations

This repository contains the simulation code used to generate the figures in the paper Dandelion++: Lightweight Cryptocurrency Networking with Formal Anonymity Guarantees (2018). These simulations compare Dandelion spreading (and several of its variants) to diffusion on different types of graphs.

To reproduce our experiments, you will need to set the appropriate parameters in config_random_regular.py, and then run:

python random_regular_propagation.py

The appropriate parameter configurations for the different figures are listed below. Unless stated otherwise, the listed code should be pasted into the appropriate part of the config file (config_random_regular.py).

Dependencies:

• Python 2.7
• NetworkX 2.2

Figure 3: Average precision as a function of p for random, directed, d-regular graphs

ds = [1,2,3]

# graph type
sim_graph = graph_lib.RegGraphGen
sim_graph_params = {}

# simulation settings
sim_settings['first_spy_dand_per_tx'] = (sim_lib.FirstSpyLineSimulator, 
										   {'p_and_r':True, 'edgebased':0})
sim_settings['max_weight_dand'] = (sim_lib.MaxWeightLineSimulator, {'p_and_r':True})

Figure 5: Recall vs # of transactions per node in random 4-regular graphs

Analogous data for Figure 5 can be produced with the scripts in the fingerprint-d2 directory under branch 'routing-simulations'

Figure 6: First-spy precision for 4-regular graphs for various routing schemes.

ds = [2] 

# graph type
sim_graph = graph_lib.RegGraphGen
sim_graph_params = {}

# simulation settings
sim_settings['first_spy_dand_per_tx'] = 
		(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'edgebased':0})
sim_settings['first_spy_dand_per_edge'] = 
		(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'edgebased':1})
sim_settings['first_spy_dand_all_to_one'] = 
		(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'edgebased':2})
sim_settings['first_spy_dand_one_to_one'] = 
		(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'edgebased':3})
sim_settings['first_spy_diffusion'] = (sim_lib.FirstSpyDiffusionSimulator, {})

Figure 7: Average precision for exact 4-regular vs. quasi-4-regular

You will have to run the code twice, with difference parameter settings.

quasi-4-regular:

ds = [2]

# graph type
sim_graph = graph_lib.QuasiRegGraphGen
sim_graph_params = {'d_anon':2}

# simulation settings
sim_settings['first_spy_dand_per_tx'] = 
	(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'edgebased':0})
sim_settings['max_weight_dand'] = 
	(sim_lib.MaxWeightLineSimulator, {'p_and_r':True})

exact 4-regular:

ds = [2]

# graph type
sim_graph = graph_lib.RegGraphGen
sim_graph_params = {}

# simulation settings
sim_settings['first_spy_dand_per_tx'] = 
	(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'edgebased':0})
sim_settings['max_weight_dand'] = 
	(sim_lib.MaxWeightLineSimulator, {'p_and_r':True})

Figure 8: Honest-but-curious spies obey graph construction protocol

# main p2p graph degree
ds = [BTC_GRAPH_OUT_DEGREE]

# graph type
sim_graph = graph_lib.QuasiRegGraphGenSpies
sim_graph_params = {'d_anon':2}

# simulation settings
sim_settings['first_spy_dand_q_0_spies_misbehave'] = 
	(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'q':0})
sim_settings['first_spy_dand_q_0_25_spies_misbehave'] = 
	(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'q':0.25})
sim_settings['first_spy_dand_q_0_5_spies_misbehave'] = 
	(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'q':0.5})

Figure 9: Malicious spies make outbound connections to every node

# main p2p graph degree
ds = [BTC_GRAPH_OUT_DEGREE]

# graph type
sim_graph = graph_lib.QuasiRegGraphGenSpiesOutbound
sim_graph_params = {'d_anon':2}

# simulation settings
sim_settings['first_spy_dand_q_0_spies_misbehave'] = 
	(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'q':0})
sim_settings['first_spy_dand_q_0_25_spies_misbehave'] = 
	(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'q':0.25})
sim_settings['first_spy_dand_q_0_5_spies_misbehave'] = 
	(sim_lib.FirstSpyLineSimulator, {'p_and_r':True, 'q':0.5})

Figures 14 and 15:

Enter branch 'routing-simulations'. The data for Figure 14 can be produced with the directories ending in -u, and the data for Figure 15 can be produced with the directories ending in -k