Waldo: A Private Time-Series Database from Function Secret Sharing

This implementation accompanies our paper by Emma Dauterman, Mayank Rathee, Raluca Ada Popa and Ion Stoica to appear at Oakland'22.

WARNING: This is an academic proof-of-concept prototype and has not received careful code review. This implementation is NOT ready for production use.

This prototype is released under the Apache v2 license (see License).

Setup

Install gRPC (tested on v1.48.1) using the instructions here.

Download the Boost library (tested on versions 1.74-1.76; can be installed on Ubuntu with sudo apt install libboost-all-dev) and Relic (tested on version 0.6.0; no presets needed while building Relic; use -DMULTI=OPENMP flag with cmake when building Relic).

To install the dependencies to run benchmarking scripts, cd scripts and run pip install -r requirements.txt.

Note that the libPSI and libOTe libraries, which build on cryptoTools, are already included in fss-core.

Building

Build libOTe

cd fss-core/libOTe
cmake . -DENABLE_RELIC=ON -DENABLE_NP=ON -DENABLE_KKRT=ON
make -j

Build libPSI

cd fss-core/libPSI
cmake . -DENABLE_RELIC=ON -DENABLE_DRRN_PSI=ON -DENABLE_KKRT_PSI=ON
make -j

Build networking code

cd network
cmake .
make
cd ..

Build Waldo

cmake .
make

Running benchmarks

Run scripts/runExperiment.py with the options for setup ('-s') and/or running the experiment ('-r').

Set the experiment config file at the top of runExperiment.py, and make sure that the experiment config file has the right IP addresses.

After you've pulled and run make at the client and servers, run python runExperiment.py at the coordinator machine (machine running the experiments, not the client or servers).

The output will be in experiments/results in the directory corresponding to the experiment and the particular time. See processed_results.dat for final benchmark numbers (look in runExperiment.py for how to parse), and you can see logs for individual executions in the directory for each parameter setting.

Testing locally

To run the entire system locally, start server X as ./build/bin/query_server config/serverX.config. Start the client with ./build/bin/bench config/client.config. Alternatively, start the client with ./build/bin/correctness_tests to run the correctness tests. Modify the parameters in the corresponding config files to run with different settings (e.g. number of cores, malicious security). Make sure to start the servers within 10 seconds of each other and wait until each has printed "DONE WITH SETUP" before starting the client (this means initialization has completed).

The following unit tests can also be run locally:

./build/bin/AggTreeUnitTest
./build/bin/DCFTableICUnitTest
./build/bin/DCFTableParallelICUnitTest
./build/bin/DCFTableUnitTest
./build/bin/DCFTableParallelUnitTest
./build/bin/DCF_unit_test
./build/bin/DPFTableUnitTest
./build/bin/DPF_unit_test

Limitations

The rollup functionality suggested as an extension for the WaldoTree construction in the paper is not fully implemented. Also, the node values for WaldoTree are directly returned without aggregating by the user-defined aggregation function.

Acknowledgements

Thanks to Natacha Crooks and Vivian Fang for contributing to the framework used for the benchmarking in scripts/.

ucbrise / waldo