华为Mate 70 Pro多实例负载均衡配置：ThinkPad X230上的AI推理加速实战（全网最详细教程）

请求入口（ThinkPad X230）
        ↓
   Nginx（负载均衡层）
        ↓
  ┌─────┼─────┐
  ↓     ↓     ↓
Mate70Pro1  Mate70Pro2  Mate70Pro3

adb connect 192.168.1.101:5555
adb connect 192.168.1.102:5555
adb connect 192.168.1.103:5555

adb devices
# 输出应显示三台设备的序列号与连接状态

upstream mate70_inference {
    least_conn;

    server 192.168.1.101:8080 weight=3 max_fails=3 fail_timeout=30s;
    server 192.168.1.102:8080 weight=3 max_fails=3 fail_timeout=30s;
    server 192.168.1.103:8080 weight=2 max_fails=3 fail_timeout=30s;

    keepalive 32;
}

server {
    listen 80;
    server_name lb.mate70.local;

    location / {
        proxy_pass http://mate70_inference;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_connect_timeout 5s;
        proxy_read_timeout 60s;
        proxy_next_upstream error timeout http_502;
    }

    location /health {
        access_log off;
        return 200 "OK\n";
        add_header Content-Type text/plain;
    }

#!/bin/bash
for ip in 192.168.1.101 192.168.1.102 192.168.1.103; do
    if curl -s -o /dev/null -w "%{http_code}" http://$ip:8080/health | grep -q "200"; then
        echo "$ip OK"
    else
        echo "$ip FAIL"
    fi
done

*/10 * * * * /opt/check_mate70.sh >> /var/log/mate70_health.log 2>&1

sudo nginx -t && sudo systemctl reload nginx

# 使用wrk进行基准测试
wrk -t4 -c100 -d60s http://lb.mate70.local/health

# 在各Mate 70 Pro上执行，统计接收到的请求数
adb -s 192.168.1.101 shell "netstat -an | grep :8080 | wc -l"

最大QPS ≈ 单节点QPS × 节点数 × 调度效率系数

upstream mate70_inference {
    least_conn;
    server 192.168.1.101:8080 weight=3 max_fails=3 fail_timeout=30s;
    server 192.168.1.102:8080 weight=3 max_fails=3 fail_timeout=30s;
    server 192.168.1.103:8080 weight=2 max_fails=3 fail_timeout=30s;
    keepalive 64;  # 提升至64
}

#!/bin/bash
# 连续3次健康检查失败，降低权重
# 连续5次健康检查成功，恢复权重

stream {
    upstream mate70_inference_tcp {
        least_conn;
        server 192.168.1.101:8081 weight=3;
        server 192.168.1.102:8081 weight=3;
        server 192.168.1.103:8081 weight=2;
    }

    server {
        listen 8081;
        proxy_pass mate70_inference_tcp;
    }