Image building performance vs Docker

[suptejas]

Hi, I was doing some research about Podman, and came across the following few sentences:

Podman provides faster startup thanks to its daemon-less structure. At any point in time, the host system stays protected. However, when it comes to overall building, Docker is a winner. The building part relies primarily on image building. Since Docker has native image building, it can build at least six times faster than Podman.

I am interested in the building part - is the above statement still true and are there any plans to improve build times by using a native image building system if this hasn't already been built out?

[flouthoc]

Hi @suptejas , I am not sure if I can totally agree with the statement above, both docker and podman allows building images in different configuration like with different storage drivers, rootless vs rootful etc. I am not at all sure if there is any official bench-marking which has been done to compare build performance if different settings and environments.

However I do agree that there are some design differences between podman and docker which are intentional for the use-case which podman provides, other than that I am not sure if there is any official bench-marking which has been done between buildah and buildkit which is used by podman and docker respectively. @containers/podman-maintainers Might be able to confirm more about if any such comparisons were done in past.

build times by using a native image building system

Podman uses buildah natively. I am not sure what do you mean by using a native image building system.

[vrothberg]

"at least six times faster" feels extremely generalized. I am unaware of that rule of thumb and assume the authors used fuse-overlayfs and did not compare root vs root and rootless vs rootless. In such a scenario, I would not expect serious performance differences with either tool. buildkit, however, seems to have some great performance boosts in certain scenarios.

[afbjorklund]

Docker has changed their "native building" to use BuildKit instead. Which improved the performance.

https://docs.docker.com/engine/release-notes/23.0/#2300

Set Buildx and BuildKit as the default builder on Linux. moby/moby # 43992

• Alias docker build to docker buildx build. docker/cli # 3314

[rhatdan]

The only way I see that they could improve performance would be with caching of previous builds or in handling of parallel entries in the Containerfile. @flouthoc has been adding a lot of these caching mechanisms to Buildah. If we had specific workloads that showed the difference in speed, then we could work to improve Buildah caching.

@nalind any other ideas?