FreeBSD Network Autotuner

Purpose

Dynamically adjusts the TCP stack, congestion control algorithm, buffers, and queues based on real-time network metrics such as latency, packet loss, jitter, throughput, queue drops, out-of-order segments, FIN-WAIT-1/FIN-WAIT-2 sockets, and NIC speed. It is designed to optimize FreeBSD networking performance in diverse and changing conditions.

---

Usage

sudo fbsd-net-autotuner [-d] [-l]

Options

• -d → Add date and time to the event log.
• -l → Log events to the file /var/log/fbsd-net-autotuner.log.

---

Configuration Variables

You can override these defaults by exporting environment variables before running the script:

Variable	Default	Description
TARGET_HOST	auto	Host used for ping measurements. Selected dynamically from public resolvers (1.1.1.1, 1.0.0.1, 8.8.8.8, 8.8.4.4, 9.9.9.9) based on lowest RTT.
IFACE	auto	Network interface to monitor. Auto-detected from default route or falls back to the first interface.
PING_COUNT	10	Number of pings per round
INTERVAL_SEC	60	Interval between measurements (sec)

---

Features

• Auto-selects the most responsive TARGET_HOST from a resilient list of public resolvers.
• Detects NIC speed robustly (validates numeric values, defaults to 1000 Mbps).
• Classifies network conditions into tiers: very_slow, slow, medium, fast, very_fast.
• Dynamically selects TCP congestion control algorithm and stack:
  • Algorithms: cubic, chd, htcp, dctcp, cdg, vegas.
  • Stacks: freebsd, bbr, rack.
• Adjusts:
  • Buffer sizes (recv/send, maxsockbuf).
  • ECN (Explicit Congestion Notification).
  • Keepalive.
  • Delayed ACK (tier-based with jitter override).
  • SACK (Selective ACK).
  • Initial congestion window (initcwnd_segments).
  • TSO (toggle based on kernel drops).
  • ABC (net.inet.tcp.abc_l_var) → controls cwnd growth aggressiveness.
  • Reassembly queue (net.inet.tcp.reass.maxqueuelen) → adapts to out-of-order segments.
  • Enables net.inet.tcp.fast_finwait2_recycle only if FIN-WAIT-2 exceeds 5% of ESTABLISHED.
• Monitors:
  • RTT (latency).
  • Packet loss.
  • Jitter.
  • Throughput.
  • Kernel intr_queue_drops.
  • Out-of-order segments.
  • FIN-WAIT-1 sockets (diagnostic only, logged).
  • FIN-WAIT-2 sockets (dynamic tuning).
• Tunes mbuf clusters and jumbo buffers based on NIC speed (1G, 10G, 40G).
• Tunes UDP receive buffer size based on NIC speed.
• Applies changes immediately via sysctl and persists them in /etc/sysctl.conf.
• Configures ISR tunables (maxthreads, bindthreads, defaultqlimit, maxqlimit) via /boot/loader.conf (requires reboot).
• Adds automatic comments with timestamp in /etc/sysctl.conf and /boot/loader.conf to mark lines generated by the autotuner for traceability.
• Flexible logging:
  • Terminal only.
  • Terminal + date.
  • File logging.

---

Network Tier Classification

The script evaluates RTT (latency), packet loss, throughput, and also jitter/out-of-order to classify the connection:

Tier	Conditions (examples)	Behavior
very_slow	RTT ≥ 250ms or loss ≥ 5%	Conservative buffers, Vegas/CHD
slow	RTT ≥ 150ms or loss ≥ 2% or throughput < 2 Mbps	Larger buffers, Vegas/HTCP
medium	RTT ≥ 80ms or throughput < 20 Mbps	Balanced buffers, HTCP/Cubic
fast	RTT ≥ 30ms or throughput < 200 Mbps	Aggressive buffers, DCTCP/HTCP
very_fast	RTT < 30ms and throughput ≥ 1 Gbps	Maximum buffers, BBR/RACK

---

Requirements

• FreeBSD system with root privileges.
• Utilities: sysctl, ping, netstat.
• Reboot required for loader.conf tunables (net.isr.*).

---

Monitored Metrics

The net-autotuner continuously measures and logs the following metrics to guide dynamic tuning decisions:

Metric	Source/Method	Purpose / Usage
RTT (latency)	ping statistics	Classify network tier (very_slow → very_fast).
Packet loss	ping statistics	Detect unstable links, adjust buffers and congestion control.
Jitter	Min/Max RTT difference	Override delayed ACK when jitter is high.
Throughput	netstat -I <iface> bytes per second	Classify tier, estimate Bandwidth-Delay Product (BDP).
Intr queue drops	sysctl net.inet.ip.intr_queue_drops	Increase intr_queue_maxlen, toggle TSO if persistent.
Out-of-order segs	netstat -s	Adjust TCP reassembly queue length (reass.maxqueuelen).
FIN-WAIT-1 conns	`netstat -an	grep FIN_WAIT_1`
FIN-WAIT-2 conns	`netstat -an	grep FIN_WAIT_2`
ESTABLISHED conns	`netstat -an	grep ESTABLISHED`

---

Notes

• The script runs in an infinite loop. Stop it with Ctrl+C.
• Ensure /etc/sysctl.conf and /boot/loader.conf is writable for persistence.
• Ensure that the congestion control algorithms (cubic, chd, htcp, dctcp, cdg, vegas) can load.
• Ensure that the TCP stacks (freebsd, bbr, rack) can load.
• delayed_ack tier setting may be overridden by jitter detection.
• abc_l_var and reass.maxqueuelen are tuned dynamically based on stability and out-of-order metrics.
• FIN-WAIT-1: monitored only, useful for diagnostics, no sysctl tuning available.
• FIN-WAIT-2: dynamically managed; fast_finwait2_recycle enabled only if >5% of ESTABLISHED.
• All metrics are logged in extended output for traceability and performance analysis.

---

Example

Run with logging to file and timestamps:

sudo fbsd-net-autotuner -d -l

---

Installation

Clone the repository and run the installer:

git clone https://github.com/geobarrod/FBSD-Net-Autotuner.git
cd FBSD-Net-Autotuner
sudo make install

---

Uninstallation

To remove the fbsd-net-autotuner and service:

cd FBSD-Net-Autotuner
sudo make uninstall

---

Changelog

v1.0 — 2025-12-05

• Initial release of FreeBSD Network Autotuner.

v1.1 — 2025-12-10

• Added BDP estimation and buffer tuning.

v1.2 — 2025-12-15

• Added jitter measurement and override for delayed ACK.

v1.3 — 2025-12-20

• Added out-of-order segment measurement and dynamic tuning for ABC and reassembly queue.

v1.4 — 2025-12-25

• Refactored loss_pct extraction.
• Improved persistence logic in set_sysctl_if_needed.

v1.5 — 2026-01-09

• Automatic selection of TARGET_HOST.
• Robust NIC speed detection.
• Mbuf cluster and UDP buffer tuning.
• ISR tuning via /boot/loader.conf.

v1.6 — 2026-01-11

• Added monitoring of FIN-WAIT-1 sockets as a diagnostic metric (logged, no dynamic tuning).
• Integrated dynamic tuning for FIN-WAIT-2 sockets:
  • Uses net.inet.tcp.fast_finwait2_recycle.
  • Enabled only if FIN-WAIT-2 exceeds 5% of ESTABLISHED connections.
  • Restores default when ratio ≤5%.
  • Idempotent logging: changes are logged only when state transitions occur.
• Extended log output to include finwait1_conns, finwait2_conns, and est_conns.

v1.7 — 2026-01-14

• Introduced adaptive congestion control selection:
  • Chooses CC algorithm dynamically (Vegas, HTCP, DCTCP, CDG, CHD/HD, Cubic) based on RTT, loss, jitter, and throughput.
• Introduced adaptive TCP stack selection:
  • Chooses stack dynamically (BBR, RACK, FreeBSD) based on throughput, RTT, loss, and jitter.