Fetch the repository succeeded.
XDP stack is listed below including PCT technologies layer which provides sandbox, TEE, FL and MPC components to support a variety of algorithms and applications. XSCE is the MPC framework component integrated in XDP platform.
XSCE (XDP Secure Computing Engine) is a unified MPC framework providing most frequently used algorithms and can be easily integrated by application software.
For current phase, XSCE provide PSI&PIR algorithms to open source, and more algorithms will come soon in the next few months.
The XSCE has been tested on Linux. There is one dependency on libOTe. The lowest dependencies of build and run environment: Ubuntu 20, c++ 17, cmake 3.20, python3, instruction set: aes sse2 sse3 ssse3 sse4.1 sse4.2 avx avx2 avx256 avx512f pclmul bmi2 adx mmx.
The XSCE can be built as:
./build.py libote spdz xsce
./build.py libote command will build libOTe library and xsce will use the library in ./third_party/libOTe_libs, otherwise xsce will use the library in the default directory which is /usr/local.
./build.py spdz command will build MP-SPDZ library and xsce will use the library in ./lib, otherwise xsce will use the library in the default directory which is /usr/local.
The libraries are libtoolkits.a, libcommon.a, libPSI.a, libPIR.a and libMP-SPDZ.a, will be located in the ./lib directory, and the main executable with examples is pir/psi/arithmetic and is located in the build directory, eg build/bin/pir, build/bin/psi, build/bin/arithmetic.
After build the XSCE, the libraries and headers for secondary developing can be installed as following:
./build.py install install the headers and libs to the default path: /usr/local
or
./build.py install=/xx/yy install the headers and libs to the configuration path: /xx/yy
The XSCE provides docker file. Executing the following commands can build a compiled and installed xsce environment, the libraries and header files will be located in /usr/local.
./build.py docker=images
After build the docker images, you can run the docker and git clone the source code, then build the examples in the docker by ./build.py xsce.
By default, sudo is not used. If installation requires sudo access, then add sudo. See ./build.py help for full details.
After building project, use the following command in the project root directory to run psi example:
build/bin/psi
After building project, use the following command in the project root directory to run pir example:
build/bin/pir
or
1. server party: build/bin/pir -r 0
2. client party: build/bin/pir -r 1
After building project, first create the test folder, then copy the executable program and circuits into it.
mkdir test_dir
cp -rf XSCE_PATH/build/bin/arithmetic XSCE_PATH/src/arithmetic/Programs test_dir/
In addition, if you want to run 3-party algorithms, you may need to set up ssl for security as we use honest majority protocol(Shamir), then copy the certificates and keys into the test directory.
cd XSCE_PATH/third_party/MP-SPDZ
./Scripts/setup-ssl.sh [<number of parties> <ssl_dir>]
If it failed for `c_rehash: command not found`, change workdirectory to <ssl_dir> and execute shell command `for file in *.pem; do ln -s "$file" "$(openssl x509 -hash -noout -in "$file")".0; done` for instead.
mkdir test_dir/Player-Data
cp XSCE_PATH/third_party/MP-SPDZ/ssl_dir/* test_dir/Player-Data/
cd test_dir
Finally, use the following command in the project root directory to run arithmetic example(e.g., 2-party addition):
1. party0: ./arithmetic -c 2 -r 0 -p 127.0.0.1:7878 add2
2. party1: ./arithmetic -c 2 -r 1 -p 127.0.0.1:7878 add2
After building project, copy the mp-spdz circuits to root directory.
cp -rf XSCE_PATH/src/arithmetic/Programs XSCE_PATH/
Finally, use the following command in the project root directory to run logistic regression example:
1. party0: build/bin/lr -r 0 -alg 1
2. party1: build/bin/lr -r 1 -alg 1
PSI algorithm is implemented according to IACR paper “Private Set Intersection in the Internet Setting From Lightweight Oblivious PRF”: https://eprint.iacr.org/2020/729
PIR algorithm is implemented as below: Input/output: client party uses single or multiple ids to query and get result from server, server party uses multiple string data for client to query each of which contains id field value.
1. client uses ids to run psi algorithm with server, and get the server indexes of matched ids in server party.
2. server encrypts each string data with random keys which are oprf values generated in PSI algorithm(for intersection elements, their oprf values are the same) and send them to client.
3. client uses local oprf values of matched elements to decrypt ciphertext data to plaintext string data.
The arithmetic algorithm is implemented based on MP-SPDZ, the circuits have been compiled and optimized. And XSCE implements a C++ wrapper of the SPDZ high-level interface, which can be used to do computation involving the adding, multiplying, etc. The following algorithms are supported:
a+b)a*b)a<b)Given two arrays, to find the median-value)Given two arrays, to compute the variance)Given two arrays, to compute the inner product)a+b+c)a*b*c)Given three arrays, to find the median-value)Given three arrays, to compute the variance)The 2-party algorithms are as follows：
EXPORT_SYM int runAdd2(SPDZAlg *spdzalg); //2-party additionEXPORT_SYM int runMul2(SPDZAlg *spdzalg); //2-party multiplicationEXPORT_SYM int runCmp2(SPDZAlg *spdzalg); //2-party comparisonEXPORT_SYM int runVar2(SPDZAlg *spdzalg); //2-party varianceEXPORT_SYM int runMid2(SPDZAlg *spdzalg); //2-party median-valueEXPORT_SYM int runInnerProd2(SPDZAlg *spdzAlg); //2-party inner-productThe 3-party algorithms are as follows：
EXPORT_SYM int runAdd3(SPDZAlg *spdzalg); //3-party addition
EXPORT_SYM int runMul3(SPDZAlg *spdzalg); //3-party multiplication
EXPORT_SYM int runVar3(SPDZAlg *spdzalg); //3-party variance
EXPORT_SYM int runMid3(SPDZAlg *spdzalg); //3-party median-value
logistic regression (vertical mode)
Logistic regression (vertical mode) algorithm is implemented as below:
Input/output：
1. training data: server provide sample data including label value and feature value, client provide correspondent sample data only including feature value(should be aligned with server);
2. traning parameters: iteration number, traning batch size, learning rate.
More algorithms will be supported soon.
This project has been placed in the public domain and/or apache2.0 license.
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
1. Open source ecosystem
2. Collaboration, People, Software
3. Evaluation model
马建仓 AI 助手