authenticated-browsing-eval

Evaluation artifacts for AuthBrowsing: the scripts, configs and patches we used to benchmark the DPDK validator on our testbed. The validator itself lives in a separate repo, alongside the load generator.

Layout

• code/: wrapper scripts that launch the validator on a host (NUMA-aware lcore pinning, NIC binding, MoonGen patches).
• experiments/: one folder per experiment, each with the same layout:
  • throughput/: packet rate and core count sweeps against the validator.
  • e2e-tcp/: end-to-end HTTP latency and req/s with an HTTP server behind an AB validator.
  • replay/: how the validator handles replays mixed into legitimate traffic.
  • triage/: how the validator handles attack traffic (forged MACs mixed with legitimate ones).

Each experiment folder has a run.sh, a setup.sh, a global-variables.yml pinning the upstream commits we tested against, and a plotting script that turns results/ into the figures.

Running

Running the experiments is easiest on a pos testbed (or something close to it). The experiments require, at a minimum, two DPDK-capable hosts with a NIC each. From an experiment folder:

./setup.sh   # provisions hosts, builds AuthBrowsing-dpdk + MoonGen
./run.sh     # sweeps the loop variables, drops raw output in results/
./plot-*.py  # turns results/ into a figure

rerun.sh runs again on existing allocations without re-provisioning, so you don't wait for reboots.

Notes

• The MoonGen patches under code/moongen-patches/ are needed for the rate control we use. Copy them on top of the pinned commit.