test environment: server, client: vultr Frankfurt 40USD 8G RAM 4 vCPU speedtest node: spOnsor="Interoute VDC" id="10268" at New York SSR setting (aes-128-cfb, auth_aes128_md5, plain)
haproxy - LKL (SSR) download 185Mbps cpu usage haproxy: 117% cpu usage python: 50% upload 50Mbps cpu usage haproxy: 40% cpu usage python: 20%
python SSR - LKL download 140Mbps cpu usage python: 149% upload 24Mbps cpu usage python: 42%
SSR native download cpu usage: 92% 450Mbps upload cpu usage: 30% 60Mbps
LKL 和 UML 的 bottleneck 是 CPU LKL 也被封的 特是 LKL 吃 CPU 也不是一般的少 --