Xray Speed Testing

A self-contained benchmark for measuring Xray throughput using VLESS + REALITY + XTLS-Vision.

How it works

Three Docker containers on an isolated bridge network:

• xray-server — Xray running as a VLESS/REALITY server
• tester — Xray client (SOCKS5) + Go program that sends data through the proxy and measures speed
• sink — TCP sink (nc -lkp 8080 > /dev/null) that discards all incoming data

The tester connects to the sink via the Xray proxy tunnel, sends a configurable amount of random data, and reports throughput.

Usage

Pull the pre-built tester image and start all services:

docker compose pull && docker compose up

Results are printed to the tester container logs when the test finishes.

To follow the output in real time:

docker compose up && docker logs -f tester

Configuration

Environment variables for the tester service in docker-compose.yml:

Variable	Default	Description
DATA_TO_SEND_MB	102400	Total data to send, in megabytes
BUFFER_SIZE	65536	Write buffer size, in bytes

Important: use a separate machine for realistic results

This setup runs the Xray client and Xray server on the same host, sharing CPU resources. In production, client and server run on separate machines with dedicated CPUs.

Because VLESS/REALITY/XTLS-Vision is CPU-intensive on both sides, co-locating them causes CPU contention and underestimates real-world throughput.

For accurate absolute numbers, run xray-server (and sink) on a remote machine and point the tester's TARGET_HOST / xray-server address at it. The current setup is still useful for comparing configurations against each other, since the bias affects all runs equally.

Project structure

.
├── docker-compose.yml
├── xray-server/
│   └── config.json       # Xray server config (VLESS inbound, REALITY)
└── tester/
    ├── Dockerfile         # Builds Go binary + downloads Xray client
    ├── client-config.json # Xray client config (SOCKS5 inbound, VLESS outbound)
    └── speedtest.go       # Go speed test program