Skip to main content
These numbers come from one run on 2026-07-24, on a NixOS workstation with Podman 5.8.2 and warm caches. They predate v0.1.0, which moved the control plane to an unprivileged account. Treat them as a rough guide. Each example app was deployed three ways:
  • Russel container: russel deploy, rootless container.
  • Russel microVM: russel deploy with type = "microvm".
  • Plain Podman: podman build from the example’s Dockerfile, then podman run.

​
Deploy to first response

Time from starting the deploy until the app answered an HTTP request, including the build. Russel’s Nix builds were cached; Podman rebuilt its image. Most of Podman’s time is the image build. With both sides cached, the gap comes down to the start-up numbers below.

​
Start-up only

Time from “the build exists” to “the app answers”, without the build: Plain Podman usually starts fastest. A Russel container adds 100 to 300 ms for preparing its root folder and checking that the app really answers. A microVM adds a kernel boot on top. Containers were capped at 256 MB of memory to match the microVM default.

​
Where a container deploy spends its time

A typical cached deploy of basic-http:

​
Run it yourself

The benchmark scripts are in the repo root and need a checkout, Nix, rootless Podman, and Podman or Docker for the Dockerfile baseline. The microVM runs also need KVM and root, and are skipped without Russel’s kernel.
--cold clears the caches first. The scripts take over /var/lib/russel while they run, so never run them on a server with real services.