SPU执行PSI实践

[yyf511]

安装环境

docker pull secretflow/secretflow-anolis8:latest

启动2个实验容器

docker run -d -ti --net="host" --name=lianxi --shm-size=10.24gb secretflow/secretflow-anolis8:latest bash
docker run -d -ti --net="host" --name=lianxi2 --shm-size=10.24gb secretflow/secretflow-anolis8:latest bash

模拟alice、bob两方

准备spu的psi代码

# -*- coding: UTF-8 -*-
"""
@Name: psi.py
@Auth: yaoyf
@Date: 2024/3/25-16:57
@Desc: aa
@Version: 0.1
"""

# system lib


# third lib
import os

import spu
import secretflow as sf
import argparse

# user lib


if __name__ == '__main__':
    cluster_config = {
        "parties": {
            "alice": {
                "address": "10.1.251.74:9301",  # proxy的ip
                "listen_addr": "10.1.251.74:9301",
            },
            "bob": {
                "address": "10.1.251.75:9301",  # porxy的ip
                "listen_addr": "10.1.251.75:9301",
            },
        },
        "self_party": "alice",
    }
    spu_config = {
        "nodes": [
            {"party": "alice", "id": "alice", "address": "10.1.251.75:9302"},   # spu的ip
            {"party": "bob", "id": "bob", "address": "10.1.251.75:9302"},
        ],
        "runtime_config": {
            "protocol": spu.spu_pb2.SEMI2K,
            "field": spu.spu_pb2.FM128,
        },
    }

    sf.init(address="10.1.251.74:9001",   # ray集群
            cluster_config=cluster_config,
            ignore_reinit_error=True,
            cross_silo_comm_backend="brpc_link",
            cross_silo_comm_options={"connect_retry_times": 60,
                                     "connect_retry_interval_ms": 5000})
    alice = sf.PYU("alice")
    bob = sf.PYU("bob")
    spu = sf.SPU(spu_config,
                 link_desc={"connect_retry_times": 60,
                            "connect_retry_interval_ms": 5000})
    input_path = {
        alice: "/opt/alice.csv",
        bob: "/opt/bob.csv",
    }
    output_path = {
        alice: "/opt/alice_psi.csv",
        bob: "/opt/bob_psi.csv",
    }
    select_keys = {
        alice: "id",
        bob: "id",
    }
    spu.psi_csv(key=select_keys,
                input_path=input_path,
                output_path=output_path,
                receiver='bob',  # 可以哪一方可以获得交集
                protocol="ECDH_PSI_2PC",
                curve_type="CURVE_FOURQ",
                sort=False,
                broadcast_result=False,
                precheck_input=False)

启动ray集群

docker exec -ti lianxi bash
ray start --head --node-ip-address=10.1.251.74 --port=9001 --include-dashboard=True --disable-usage-stats

2方相同的操作，相同的代码，只是连接的ray集群和party换成执行方

制作数据集

在每一方制作一个只有id列的数据集即可。

执行求交

python3.8 psi.py

alice方

bob方

结果展示

alice

bob

交集