Jamming Simulator

This simulator implements outgoing reputation for a network of nodes and performs a sink attack against a target node in the network. In this attack, a malicious node will:

• Open up channels and forward payments as usual to bootstrap its reputation with a target node.
• Slow jam the target's peers's general resources, so that all HTLCs must be accountable to reach the target.
• Hold any accountable HTLCs from the target with the goal of ruining its reputation with its peers.

Requirements

To run the simulation you will need:

• sim_file.json: a json file describing the network that is being simulated, including channels that the attacking node has created to draw in traffic. An example is available here.
• bootstrap.csv: a csv containing projected forwards for the network including that attacker's channels.
• peacetime.csv: a csv containing projected forwards for the network excluding the attacker's channels.

Projections are generated using a sim-ln branch that runs on a fully simulated network and outputs forwarding records for each node in the network.

To run the simulator with the above files in the current directory:

make install
ln-simln-jamming --target-alias {alias string} --attacker-alias {alias string}

There are various ways that the simulator can be customized, see cargo run -- --help for details. A script is also provided to run a set of simulations with different attacker bootstrap values in ./run_simulations.sh.

Attack Implementation

On startup, the simulator will use the bootstrap-file to set up reputation scores for the network:

• Honest nodes will have payments replayed for the full reputation_window that the simulator is configured with (default: 6 months, equal to revenue-window-seconds reputation-multiplier).
• The attacking node will have payments replayed for the attacker-bootstrap, which must be <= reputation_window.

This sets up a starting point where the attacker has been peacefully forwarding payments for the bootstrap period provided, and all honest channels in the network have been around for at least the reputation_window.

The simulation will then start, implementing the attack as follows:

• General jamming the target's peers, so that only accountable HTLCs reach the target node.
• The attacking node will hold HTLCs for the maximum allowable period before risking a force close, then fail them back.

To compare revenue under attack, the forwards in the peacetime.csv projection will be replayed as the simulation executes.

The simulation will shut down if:

• The target node's revenue in the simulation dips below its projected peacetime revenue (it has suffered revenue loss).
• The attacker no longer has reputation with the target (it can't continue its attack).
• The simulator has run out of peacetime forward to replay to compare revenue.

Shortcomings

• This simulator relies on sim-ln to generate payment flows and projections, so traffic only represents our best guess at how payments flow in the network.
• Payments are generated with a fixed seed, but this is not perfectly deterministic in sim-ln.
• The simulator implements "just in time" upgradable accountability, bumping up the accountability signal of a htlc only if it is required.
• At present the simulator only expects one channel between the target and attacking node.