TCP Brutal (Enhanced Version)

Overview

TCP Brutal is an enhanced version of the original TCP Brutal, offering a user-friendly, maintainable, and extensible TCP congestion control kernel module. It introduces advanced runtime configuration and monitoring features, making it suitable for operators, developers, and researchers.

Key Differences from the Original Brutal:

• Runtime Parameter Tuning: Supports hot-updating key algorithm parameters via both sysfs and procfs, without requiring module reload.
• Machine-Readable Monitoring: Exposes detailed, machine-friendly statistics under /proc/brutal_stats for easy automated collection.
• Sysfs Integration: All tunable parameters are visible and writable in /sys/module/brutal/parameters/.
• Better Logging & Robustness: Enhanced error logging, boundary checks, and documentation for easier debugging and integration.
• Full Backward Compatibility: All core TCP Brutal logic is preserved; only the configuration and monitoring experience is extended.

For Users

Installation

Quick install script:

bash <(curl -fsSL https://tcp.hy2.sh/)

Manual build and load:

# Ensure kernel headers are installed (e.g. Ubuntu: apt install linux-headers-$(uname -r))
make && make load

Requires Linux kernel 4.9 or later (5.8+ recommended).
On kernels older than 5.8, IPv6 is not supported.
On kernels older than 4.13, you MUST manually enable fq pacing:
tc qdisc add dev eth0 root fq pacing

Monitoring and Tuning

• All key parameters (initial rate, cwnd_gain) can be viewed and changed at runtime:
  • Sysfs:

    cat /sys/module/brutal/parameters/param_init_rate
    echo 200000 > /sys/module/brutal/parameters/param_init_rate
    

  • Procfs (stats):

    cat /proc/brutal_stats
    

    Output is machine-readable, e.g.:

    initial_rate:125000
    initial_cwnd_gain:20
    min_pacing_rate:62500
    ...
    

• No kernel reload or module reload is needed for parameter changes.

---

For Developers

This kernel module provides the brutal TCP congestion control algorithm, which can be enabled in any TCP socket using TCP_CONGESTION:

s.setsockopt(socket.IPPROTO_TCP, TCP_CONGESTION, b"brutal")

Setting Algorithm Parameters

To set the send rate and congestion window gain at runtime (recommended: cwnd_gain 1.5x–2x, i.e. 15–20):

struct brutal_params {
    u64 rate;      // Send rate in bytes per second
    u32 cwnd_gain; // CWND gain in tenths (10=1.0)
} __packed;

import socket, struct
TCP_BRUTAL_PARAMS = 23301
rate = 2_000_000  # 2 MB/s
cwnd_gain = 15
brutal_params_value = struct.pack("QI", rate, cwnd_gain)
conn.setsockopt(socket.IPPROTO_TCP, TCP_BRUTAL_PARAMS, brutal_params_value)

Compatibility

• TCP BrutalX does not change the TCP protocol, only congestion control—works unilaterally on client or server.
• Programs must use the custom sockopt to fully leverage BrutalX (most do not by default).
• All features of the original Brutal are preserved; BrutalX simply makes the system easier to operate and monitor.

---

Example: Speed Test

The example directory contains a simple Python speed test client/server:

# Server: listens on TCP port 1234
python server.py -p 1234

# Client: connects to example.com:1234, requests 50 Mbps
python client.py -p 1234 example.com 50

---

Why choose TCP BrutalX over original Brutal?

• No more module reloads for parameter tuning—change settings on the fly!
• Easier monitoring: /proc/brutal_stats is ready for your scripts and dashboards.
• Cleaner codebase: More robust error handling, easier to maintain and extend.
• Perfect for automation and research: Integrate with your configuration management and observability stack.

---

License

GPL v3. See LICENSE.

---