Rust implementation of Shamir's secret sharing with Lambdaworks polynomials and FFT-friendly fields. Using the Stark 252 prime field from Lambdaworks.
I wrote a post about this optimization.
Currently, the project is still a WIP. I have 3 Reliable Broadcast protocols in the consensus folder. The consensus/src/sss folder contains the secret sharing protocol. The consensus/src/sss_fft file is where I am attempting to speed up the share creation step of secret sharing.
The scripts folder contains scripts to test the protocols. The testdata folder contains the configurations for the tests.
You can find the benchmarks in consensus/benches/. To run the benchmarks, use the following command:
cd consensus
cargo benchThe following are the results after running the benchmarks on my machine:
| Task | Time (µs) | % Outliers |
|---|---|---|
| Secret Generation | 11.496 - 12.480 | 7.00% |
| Secret Generation using FFT | 4.7853 - 4.8140 | 24.00% |
| Reconstructing Secret | 146.34 - 147.32 | 5.00% |
| Reconstructing Secret using FFT | 144.74 - 145.88 | 4.00% |
| Fill Evaluation at all points | 222.58 - 223.84 | 22.00% |
| Fill Evaluation at all points (FFT) | 206.96 - 211.90 | 13.00% |
- Generating shares with FFT is significantly faster (~4.8µs vs. ~12.4µs).
- Reconstruction using FFT does not provide a drastic improvement, which is expected due to interpolation.
- FFT-based fill evaluation is slightly faster.
- Kill these processes
sudo lsof -ti :7000-7015,5000 | sudo xargs kill -9- Script to test with n number of nodes (in case you lose config by switching branches)
cd testdata
mkdir hyb_16
./target/release/genconfig --NumNodes 16 --delay 10 --blocksize 100 --client_base_port 7000 --target testdata/hyb_16/ --payload 100 --out_type json --base_port 9000 --client_run_port 4000 --local true- Script to test RBC with test_msgs.txt
./scripts/test.sh testdata/hyb_16/syncer Hi false testdata/test_msgs.txt- SSS Unit Tests
cargo test