Reputation: 111
I created a cross compilation container image with build environments for different target platforms (bare metal arm core, MINGW/Windows, armhf/Raspberry Pi).
All the following was done under Manjaro Linux on an Intel NUC 8i7BEH. I was logged in as average user.
Building the image already reveals massive performance differences between docker and podman. On my rather low performance machine I see a docker build to run more than 6 times faster than the same task done with podman.
docker build - 4 min 36 sec
podman build > 25 minutes
Building my application for a single target then uses in average:
podman - 73 sec
docker - 4.9sec
A native build on the same machine needs 3 sec in average.
Starting the container image to build the app for 3 different platforms in sequence needs (average):
podman - 85 sec
docker - 13 sec
How can this massive difference in container performance be explained? And is there anything I can do to increase podman's performance, so that it gets at least into the same area as docker? I'd rather use podman than docker. I prefer the service-less setup podman provides as well as rootless containers.
Thanks Stefan
Upvotes: 11
Views: 16433
Reputation: 1939
Looks like the original poster got some help here: https://github.com/containers/podman/issues/13226
tl;dr Make sure you're using fuse-overlayfs rather than the VFS storage driver.
Upvotes: 10