This is a small, independent journal about running Linux servers on modest hardware: TCP tuning that actually moves the needle, monitoring that earns its keep, and the boring habits that keep small systems alive for years. New notes are published when there is something verified worth writing down — everything here was measured on a real machine first.
Switching the congestion control to BBR is the easy part. On small virtual machines the default socket buffer ceilings are still sized for a LAN, and a single flow across a 100 ms path stalls long before the link is full. Notes on sizing buffers from the bandwidth-delay product, pairing BBR with the fq qdisc, and verifying the change with controlled transfers instead of synthetic benchmarks.
Read the articleA 1 GB instance with no swap is one traffic spike away from the OOM killer making decisions for you. A modest swap file with a low swappiness value does not make the machine faster — it makes failure slower, observable, and survivable. How we size it, and the alerts that tell us it is being used before it becomes a problem.
Password authentication on an internet-facing server is an open invitation to every scanner on the planet. Moving to key-only login takes ten minutes; doing it without locking yourself out takes a written checklist: verify the key in a second session, keep a console path, then close the password door and prove it is closed from the outside.