Benchmarks: nyxd vs Docker vs Podman (Alpine 3.23 / QEMU)

Compare the same workload on three as similar as possible Alpine Linux 3.23 guests under QEMU. One VM runs Docker, one Podman, one nyxd + nyx. Goal: apples-to-apples numbers for startup, resource use, and steady-state behavior—not a claim of “fastest in all cases.”

Environment matrix (fill before each run)

Setting	Docker VM	Podman VM	nyxd VM
QEMU CPU model / -smp
RAM (-m)
Disk backend (qcow2 raw, cache mode)
Alpine version / kernel (uname -a)
Docker version (docker version)
Podman version (podman version)
nyxd commit / nyx version
Date / timezone

Keep CPU count, RAM, and disk identical across VMs. Prefer virtio NIC and disk. Snapshot or reinstall from the same base image between product runs.

Workload (must be identical)

1. Use the script or contract in workload/README.md (e.g. pull fixed image tag → run container with fixed flags → warm-up → measured phase).
2. Record exact commands for each stack (Dockerfile / podman run / nyx pull + nyx run) in your run notes so runs are reproducible.
3. Run the same client-side load generator from a fourth machine or the host against the published container port (same concurrency, duration, and seed if applicable).

What to measure (example rows)

Copy TEMPLATE.md into results/YYYY-MM-DD-run-N.md and fill after each trial.

Suggested metrics:

Metric	How to capture (examples)
Cold pull → first ready	time until health check passes
Daemon RSS after steady state	ps / systemd-cgtop
Container start latency	time wrapper or daemon logs
Steady CPU %	pidstat, top batch mode
Load test throughput / p99	wrk, hey, vegeta (same binary + args)
Stop / remove latency	scripted teardown

Tracing (optional)

• Enable nyxd --log-level debug for one run; archive logs under results/…/.
• Docker: journalctl -u docker or daemon debug (document flags).
• Podman: rootless vs rootful affects numbers—pick one and keep it consistent across all runs in a series.

Ethics / license

This repo is PolyForm Noncommercial for the nyxd code; benchmark numbers and your scripts in results/ are yours. Do not redistribute proprietary vendor binaries from inside this folder.

See also

• QEMU Alpine — booting Alpine 3.23 under QEMU for nyxd experiments.
• Usage — nyx commands used in the nyxd column.