Programmable Oblivious PRF & multi-party PSI

This is the implementation of our CCS 2017 paper: Practical Multi-party Private Set Intersection from Symmetric-Key Techniques[ePrint].

Evaluating on a single Intel Xeon server (2 36-cores Intel Xeon CPU E5-2699 v3 @ 2.30GHz and 256GB of RAM), ours protocol requires only 71 seconds to securely compute the intersection of 5 parties, each has 2^20-size sets, regardless of the bit length of the items.

For programmable OPRF, this code implements:

Table-based OPPRF
Polynomial-based OPPRF
BloomFilter-based OPPRF

For PSI, we implement multi-party PSI (nPSI) in augmented-semihonest model and standard semihonest model.

Installations

Required libraries

C++ compiler with C++14 support. There are several library dependencies including Boost, Miracl, NTL , and libOTe. For libOTe, it requires CPU supporting PCLMUL, AES-NI, and SSE4.1. Optional: nasm for improved SHA1 performance. Our code has been tested on both Windows (Microsoft Visual Studio) and Linux. To install the required libraries:

windows: open PowerShell, cd ./thirdparty, and .\all_win.ps1 (the script works with Visual Studio 2015. For other version, you should modify MSBuild at several places in the script.)
linux: cd ./thirdparty, and bash .\all_linux.get.

NOTE: If you meet problem with all_win.ps1 or all_linux.get which builds boost, miracl and libOTe, please follow the more manual instructions at libOTe

Building the Project

After cloning project from git,

Windows:

build cryptoTools,libOTe, and libOPRF projects in order.
add argument for bOPRFmain project (for example: -u)
run bOPRFmain project

Linux:

make (requirements: CMake, Make, g++ or similar)
for test: ./bin/frontend.exe -u

Running the code

The database is generated randomly. The outputs include the average online/offline/total runtime that displayed on the screen and output.txt.

Flags:

-u		unit test which computes PSI of 5 paries, 2 dishonestly colluding, each with set size 2^12 in semihonest setting
-n		number of parties
-p		party ID
-m		set size
-t		number of corrupted parties (in semihonest setting)
-a		run in augmented semihonest model. Table-based OPPRF is by default.
			0: Table-based; 1: POLY-seperated; 2-POLY-combined; 3-BloomFilter
-r		optimized 3PSI when r = 1

Examples:

1. Unit test:

./bin/frontend.exe -u

2. nPSI:

Compute PSI of 5 parties, 2 dishonestly colluding, each with set size 2^12 in semihonest setting

./bin/frontend.exe -n 5 -t 2 -m 12 -p 0 
& ./bin/frontend.exe -n 5 -t 2 -m 12 -p 1
& ./bin/frontend.exe -n 5 -t 2 -m 12 -p 2
& ./bin/frontend.exe -n 5 -t 2 -m 12 -p 3
& ./bin/frontend.exe -n 5 -t 2 -m 12 -p 4

Compute PSI of 5 parties, each with set size 2^12 in augmented semihonest setting with Bloom filter based OPPRF. Note that, the augmented SH protocol protects from a collusion of n-1 parties

./bin/frontend.exe -n 5 -a 3 -m 12 -p 0 
& ./bin/frontend.exe -n 5 -a 3  -m 12 -p 1
& ./bin/frontend.exe -n 5 -a 3  -m 12 -p 2
& ./bin/frontend.exe -n 5 -a 3  -m 12 -p 3
& ./bin/frontend.exe -n 5 -a 3  -m 12 -p 4

Summary

  1. git clone https://github.com/osu-crypto/MultipartyPSI.git  
  2. cd thirdparty/
  3. bash all_linux.get 
  4. cd ..
  5. cmake .
  6.  make -j
  7. ./bin/frontend.exe -n 5 -t 2 -m 12 -p 0 & ./bin/frontend.exe -n 5 -t 2 -m 12 -p 1  & ./bin/frontend.exe -n 5 -t 2 -m 12 -p 2 & ./bin/frontend.exe -n 5 -t 2 -m 12 -p 3 & ./bin/frontend.exe -n 5 -t 2 -m 12 -p 4

Help

For any questions on building or running the library, please contact Ni Trieu at trieun at oregonstate dot edu

eshf/MultipartyPSI