AnonTestLab is a local network emulator for evaluating anonymous communication designs with real relay processes, real sockets, real per-hop cryptography and configurable adversaries.
A reusable middle ground between abstract simulation and modifying a production anonymity network. Controlled enough for experiments, concrete enough to observe a running protocol.
A new anonymity mechanism can look promising for many different reasons: the protocol itself, the simulator assumptions, the traffic model, the topology, the attacker visibility, or simply the way the prototype was built. Across studies, those variables are often bundled together.
A common experimental environment makes it possible to execute, vary and challenge those ideas under the same rules.
Simulation is powerful for controlled studies, but process scheduling, socket behavior, cryptographic work and implementation timing may be modeled, while an executing protocol can expose effects that those models omit.
Testing one new routing policy, mixing defense or observer model inside a mature anonymity network can require substantial engineering before the research question itself can be isolated.
If each study builds a new harness, changes in topology, workload, timing and adversary assumptions can become entangled with the mechanism being evaluated.
Researchers should be able to change one experimental assumption at a time: routing, path splitting, cover traffic, relay mixing, cryptography, link conditions or attacker visibility. The rest of the environment should stay controlled.
And when execution matters, the measurements should be able to come from real relay processes, sockets, circuit construction and per-hop cryptography, giving timing studies an execution-based measurement layer.
AnonTestLab provides a common execution layer for asking controlled questions about anonymous communication designs before the cost of a production implementation is justified. Tor and analytical simulation remain useful for different purposes.
Change the assumption.Configuration-driven experiments with preserved seeds, metrics and reports.
Re-run the network.
Measure what changed.
Routing, traffic, mixing, shaping, cryptography, link conditions and adversary visibility are independent dimensions of the experiment.
Compare single-path and multipath behavior with multiple path-selection and traffic-splitting policies.
Vary traffic distributions and apply relay-side delay defenses without creating a separate testbed.
Circuits perform telescoping handshakes and use selectable AEAD algorithms instead of adding an artificial crypto delay.
* AES-GCM-SIV requires cryptography ≥ 42 with OpenSSL 3.2+. Standard cryptography wheels satisfy this on common platforms.
** AES-OCB3 requires OCB support in the OpenSSL build.
Make localhost links behave less like localhost while retaining a controlled, repeatable laboratory.
Make attacker capability an explicit experimental parameter, including what the observer can see.
Move beyond a single run with tooling designed for comparison, repeated seeds and experimental validity.
01ConfigureDefine network, routes, traffic, crypto and observers in YAML.02SpawnLaunch relay subprocesses on local loopback addresses.03BuildConstruct telescoping circuits with per-hop ECDHE and keys.04RunSend onion-wrapped traffic across real sockets and configured links.05MeasureExport metrics, configuration, seed and experiment report.experiment: name: multipath-study seed: 20260903 duration_s: 8 routing: paths: - strategy: bandwidth_weighted path_length: 3 - strategy: random path_length: 4 split_strategy: iid split_weights: [0.7, 0.3] merge: common_exit traffic: distribution: pareto real_rate: 6 cover_rate: 2 mixing: strategy: pool pool_interval_ms: 250 crypto: algorithm: chacha20poly1305 keyexchange: x25519
atl runExecute one experiment and collect its metrics.atl compareRun two configurations and diff every metric.atl sweepVary one field over a set of values and export the series.atl pairedRun matched seeds and classify the paired effect against an equivalence margin.atl predictInspect the model's closed-form expectations without executing a run.atl fidelityCheck whether host scheduling noise is contaminating a timing experiment.atl dashboardUse a local browser UI backed by the same experiment runner.Each question can be tested under explicit, configurable assumptions.
Does path splitting still help when the observer can see more than one leg?
Does a delay defense survive when the attacker becomes lag-aware?
How much correlation reduction is gained for the additional bandwidth cost?
When does pool or exponential mixing produce an effect that remains after host noise is accounted for?
How much of the network must an adversary observe before linkage becomes reliable?
Is the result caused by the protocol configuration, or by the machine running the experiment?
AnonTestLab is a research instrument with a deliberately bounded model. Its limitations belong in the methodology.
sessions.max_concurrent, each session starts its experiment clock when that session begins. Timing comparisons should account for this.AnonTestLab is released under the GNU General Public License v3.0 or later. That choice is intentional.
Users are free to study it, modify it, redistribute it and build on it, while the GPL preserves those freedoms for others who receive modified versions. For a research tool, that matters: experimental methods should be inspectable, implementations should be challengeable, and redistributed modified versions should preserve the same freedoms.
The license supports practical reuse while keeping the project's free-software principles explicit.
Read the GPL-3.0-or-later license ↗Clone AnonTestLab, start with the Tor-like preset, then change the routing, traffic, defenses or adversary and measure the effect.