Implementation of the Lox bridge authority system

Ian Goldberg 3fe2f00764 Update README vor 2 Jahren
Parsing-results 8ede9f32a5 Added some fixes and requested touchups from artifact review vor 2 Jahren
src b46b54e0fc cargo fmt vor 2 Jahren
tests 0e5876a939 API for adding bridges and marking them as unreachable vor 3 Jahren
Cargo.toml 231e700031 Proposed changes for artifact, run_tests_fast feature runs with 100 users vor 2 Jahren
Dockerfile 8ede9f32a5 Added some fixes and requested touchups from artifact review vor 2 Jahren
LICENSE 5269d63aba Add MIT licence vor 2 Jahren
README.md 3fe2f00764 Update README vor 2 Jahren
build-lox.sh 217a0fe383 Touch up the docker environment vor 2 Jahren
run-lox.sh 217a0fe383 Touch up the docker environment vor 2 Jahren
run_tests_fast.sh 8ede9f32a5 Added some fixes and requested touchups from artifact review vor 2 Jahren

README.md

Lox

Lox is a reputation-based bridge distribution system that provides privacy protection to users and their social graph and is open to all users. Lox is written in rust and requires cargo to test. Install Rust. We used Rust version 1.56.0.

To use the docker environment to build the package:

./build-lox.sh
./run-lox.sh

To run each of the tests used for our experimental results run:

cargo test --release -- --nocapture TESTNAME > LOGFILE

Where TESTNAME > LOGFILE is one of:

stats_test_trust_levels  > trust_levels.log
stats_test_invitations > invitations.log
stats_test_percent_blockage_migration_05 > blockage_migration05.log
stats_test_percent_blockage_migration_010 > blockage_migration010.log
stats_test_percent_blockage_migration_20 > blockage_migration20.log
stats_test_percent_blockage_migration_25 > blockage_migration25.log
stats_test_percent_blockage_migration_35 > blockage_migration35.log
stats_test_percent_blockage_migration_40 > blockage_migration40.log
stats_test_percent_blockage_migration_45 > blockage_migration45.log
stats_test_percent_blockage_migration_50 > blockage_migration50.log
stats_test_percent_blockage_migration_55 > blockage_migration55.log
stats_test_percent_blockage_migration_60 > blockage_migration60.log
stats_test_percent_blockage_migration_65 > blockage_migration65.log
stats_test_percent_blockage_migration_70 > blockage_migration70.log
stats_test_percent_blockage_migration_75 > blockage_migration75.log
stats_test_percent_blockage_migration_80 > blockage_migration80.log
stats_test_percent_blockage_migration_85 > blockage_migration85.log
stats_test_percent_blockage_migration_90 > blockage_migration90.log
stats_test_percent_blockage_migration_95 > blockage_migration95.log
stats_test_percent_blockage_migration_100 > blockage_migration100.log

Each test outputs results to the specified log file and takes approximately 20-30 hours to run. However, this can be improved by passing the fast feature. Using this feature, our tests are run for 100 users instead of 10000 users and will produce results comparable to our reported results (with larger error margins). To run individual tests with this flag run:

cargo test --release --features=fast -- --nocapture TESTNAME > LOGFILE

We have also included the scripts we used to parse the output from each of the Lox tests in the Parsing-results directory. For convenience, copy all of the output log files to the Parsing-results directory and run ./parse_data.sh. This is a python script that uses Python 3.8+ and depends on numpy, matplotlib, and pandas which can be installed with pip3.

To run all tests in fast mode, output the results to the Parsing-results directory, and generate the table (performance_stats.csv, our Table 4) and graphs (Blockage-response-time.pdf and core-users.pdf, our Figures 1 and 2) used in our paper, run:

./run_tests_fast

This takes 5–6 hours to complete on the 2011-era E7-8870 processor we used for the measurements in the paper. Newer processors will likely take less time. Note that with the fast feature, it is possible that some tests may only yield 0 or 1 data points, in which case you will see 0 for the times and/or stddevs in the resulting performance_stats.csv table for those tests.

Note that our implementation is coded such that the reachability certificate expires at 00:00 UTC. A workaround has been included in each test to pause if it is too close to this time so the request won't fail. In reality, if the bucket is still reachable, a user could simply request a new reachability token if their request fails for this reason (a new certificate should be available prior to the outdated certificate expiring).