Skip to content

macOS container runtime benchmark: VMPal vs 9 others (file I/O & build noticeably slower) #137

Open

Description

@chautruonglong

Hi team, I'm a daily VMPal user. I used an AI assistant to run and measure this benchmark of 10 container runtimes on macOS, and I'm sharing it here for the team's reference.

The main finding is that VMPal looks slow on file I/O. Volume 4K write+fsync is only ~138 IOPS, versus 2,330–19,321 for every other runtime. Bind-mount random 4K I/O is about 3× slower than Docker Desktop and about 10× slower than OrbStack. Builds are 1.4–2.2× slower than Docker Desktop, likely related to the I/O above.

Strengths: near-instant cold start (~0.8 s to first container), 53 MiB idle RAM, ~48 ms docker run, 50 containers in 0.89 s, top container→host throughput (~93 Gbit/s), and excellent amd64 emulation.

Overall scores (0–100, higher is better): VMPal 63.8, Kiri 67.5, OrbStack 57.6, Apple container 56.1, Docker Desktop 52.4, Colima 46.7, Lima 43.1, Rancher (moby) 40.4, Rancher (containerd) 36.7, Podman 33.4.

Environment: bare-metal Mac mini, Apple M6, 16 GB RAM, macOS 27.2.1; each engine run alone with 4 CPU / 8 GB where configurable; every number is the median of 5 runs (VMPal is the average of two full passes).

Full report with charts (PDF): VMPal-benchmark-report.pdf

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions