中文

XSCE

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.

build options

• build

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.

• install

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

• docker

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.

example

• psi example

After building project, use the following command in the project root directory to run psi example:

build/bin/psi

• pir example

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

• arithmetic example

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

• logistic regression example

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

Algorithms design

• PSI

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

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. 

• Arithmetic

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:

• 2-party addition(a+b)
• 2-party multiplication(a*b)
• 2-party comparison(a<b)
• 2-party median-value(Given two arrays, to find the median-value)
• 2-party variance(Given two arrays, to compute the variance)
• 2-party inner product(Given two arrays, to compute the inner product)
• 3-party addition(a+b+c)
• 3-party multiplication(a*b*c)
• 3-party median-value(Given three arrays, to find the median-value)
• 3-party variance(Given three arrays, to compute the variance)

The 2-party algorithms are as follows：

• EXPORT_SYM int runAdd2(SPDZAlg *spdzalg); //2-party addition
• EXPORT_SYM int runMul2(SPDZAlg *spdzalg); //2-party multiplication
• EXPORT_SYM int runCmp2(SPDZAlg *spdzalg); //2-party comparison
• EXPORT_SYM int runVar2(SPDZAlg *spdzalg); //2-party variance
• EXPORT_SYM int runMid2(SPDZAlg *spdzalg); //2-party median-value
• EXPORT_SYM int runInnerProd2(SPDZAlg *spdzAlg); //2-party inner-product

The 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.

1. based on gradient descent algorithm, client needs to use server label value to update its model parameters which is implemented MP-SPDZ to run a inner product calculation;
2. when training is over,server and client save their model result to specified file.

More algorithms will be supported soon.

License

This project has been placed in the public domain and/or apache2.0 license.