Northbound Notes
Field notes / 2026

Small systems,
quietly reliable.

Practical notes about Linux, networking, observability, and keeping services boring in production.

Infrastructure · 8 min read

Designing a tiny server that can recover itself

A resilient service does not need an elaborate platform. A clear boot path, narrow permissions, useful health checks, and one tested rollback are often enough.

$ systemctl is-active edge-service
active
$ sysctl net.ipv4.tcp_congestion_control
net.ipv4.tcp_congestion_control = bbr
Networking · 5 min read

Latency is a distribution, not a number

Median response time tells only half the story. Tail latency is where interactive systems begin to feel unreliable.

Operations · 4 min read

Backups are finished only after restore

A checksum confirms the archive. A rehearsal confirms the system. Keep both close to the recovery plan.