Podman or Docker CE 20.10 for best performance?

Elastic Orchestration Elastic Cloud Enterprise (ECE)
1

Hi,

I would like to know if anyone have noticed any performance issues with Redhat 8 and Podaman?

We are in the decision to either use podman or docker ce 20.10 on red hat 8 on new hw, cisco ucs 240 m7 with nvme drives for hot.

We would like to have the best performance as you can imagine.

Br/
Stefan

2

Hello.

We recently ran a number of performance tests comparing docker and podman. From what we observed the main difference can be related to that with docker it works to run with unlimited CPU in ECE under deployments while in podman it is not possible. This is, according to the support, due to how podman works with the limitations and if it's possible to change once the container is running or not.

When running with the CPU limit on under all deployments, the main difference we observe is the startup of each instance that can become quite slow. This is yet again related to the previous "function", according to support the CPU limit is off during startup of an instance but since it doesn't work to disable the CPU limit in podman the startup happends with the resources the instance has allocated.

Once an instance is up and running, we didn't have any larger differences in performance between podman and docker while running Rally tests. We were running ECE 3.6.0 and Elasticsearch 8.10.4

1 Like
3

Hi,

Thanks for the answer.

Could your tests depends on this, I found this information in another post?
ECE does not have

"Finally, I'll note that you seem to be comparing root Docker to rootless Podman, which we don't really expect to perform comparably - rootless will be slower, due to the compromises it has to make in the name of improved security. However, I think we'd expect it to be ~15% slower, not the figures you are seeing."

Podman run/build is painfully slow compared to docker · Issue #13226 · containers/podman · GitHub

Did you implement podman?
Red Hat don´t support Docker CE with our server modell so we need to consider that to.

Br/
Stefan

4

No, as I mentioned we see similar performances with docker and podman when they run with similar resources. Our discrepancies were connected to situations where docker were running without any CPU limitation.

We will probably not implement podman but go with unsupported docker instead because of the savings its related to in our environment, but that is of course a decision one has to make per implementation.

5

Hi,

We will go for 3 new allocators in test with podman to be able to compare.

Has anyone else any information to share regarding this topic?

Would be most intressting to hear your stories.

6

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.