lima-docker-lean

Raw results

Unedited output of the runs quoted in docs/en/analysis.md (RU). Apple M5, 16 GB, macOS 26.6.2, Lima 2.2.0, 2026-09-24. The instance names in the files (dev, dev-cloud) refer to the variant listed below.

File Variant What
01-iso-idle-and-load.txt ISO (alpine-lima 3.24.1) idle with FDB + Postgres, pgbench, idle after. The first FDB attempt fails (libfdb_c.so not on the loader path); the successful re-run is appended at the end.
02-cloud-untuned-empty-idle.txt cloud 3.23, stock empty idle
03-cloud-untuned-idle-and-load.txt cloud 3.23, stock idle with FDB + Postgres, pgbench, FDB, idle after
04-cloud-init_on_free-memtest.txt cloud + init_on_free=1 bench/memtest.sh: load → drop_caches → host memory pressure → stop; running/stopped are host vm_stat
05-final-template-idle-and-load.txt final minimal-docker.yaml bench/load.sh
06-iso-wakeup-snapshot.raw.txt ISO raw /proc/interrupts + per-thread context switches, two snapshots 30 s apart
07-cloud-wakeup-snapshot.raw.txt cloud, stock same
08-final-template-empty-idle.txt final template empty idle (120 s window) + one-session disk writes (180 s)

The control run for the memory experiment (without init_on_free) was done by hand right after 03, so it has no file of its own:

guest drop_caches: footprint 6158 MB before and after
memory_pressure -l warn, 40 s: footprint stays 6158 MB (vmmap: resident 6.0G, swapped_out 4.9G)
running: free=2912 active=3158 inactive=3033 wired=1972 compressor_occ=4520 stored=10298 (MB)
stopped: free=7105 active=2328 inactive=2644 wired=1964 compressor_occ=1495 stored=5255 (MB)

The per-step CPU figures for the cloud image tuning (kernel upgrade, hotplugd/syslog, ttyAMA0) come from bench/idle.sh runs whose output was not saved; they are quoted in the analysis.