VMess Protocol Test

VMess Protocol Test

What Was Tested

Add 10000 users to simulate a large number of users using the VMess protocol at the same time

Test Description

The server supports both VMess+MD5 and VMessAEAD authentication; the client connects using VMessAEAD over websocket transport, with 10000 users, alterId 0

Test Environment

Ubuntu 20.04, 4 cores, 8 GB RAM, no swap

Test Method

Client->Proxy program->Server (simulated website)

1. The client opens 3000 connections to the server through the proxy. On each connection, the client sends a random 1KB~16KB packet to the server every 2 seconds, and the server sends a random 1KB~16KB packet to the client every 1 second
2. Each connection stays open for a random 30~120 seconds and then closes; a new connection is opened immediately after each one closes, keeping the total number of active connections constant
3. Run for 10 minutes, idle for 5 minutes, run for another 5 minutes, then idle for 10 minutes, for a total test time of 30 minutes, recording CPU, memory, connection count, and network throughput

v2ray Monitoring Chart

The periodic CPU spikes occur because the VMess protocol regenerates authentication info for all users every 10 seconds, which consumes CPU. This is inherent to the VMess protocol, and the same applies to xray and soga below

xray Monitoring Chart

soga Monitoring Chart

Test Data Summary

Program	Peak Memory Usage	Memory Usage at End	Avg CPU Usage (Peak Period)	Avg CPU Usage (Idle Period)
v2ray-core(v4.45.2), 16KB connection buffer	1079MB	565MB	33.62%	1.45%
xray-core(v1.5.9), 16KB connection buffer	1084MB	553MB	31.74%	1.43%
soga(v2.7.7), default 16KB connection buffer, v2ray_reduce_memory not enabled (memory can be even lower when enabled)	604MB	276MB	21.25%	1.58%