Docker network_mode

[_Nick]

Hi all, 

According to the Jellyfin Docker install docs (https://jellyfin.org/docs/general/instal...container/), bridge is the default network mode, with host only required for DLNA (never used). However, the docker run and compose example templates use host mode, without explaining if it's performance-related.  In other words, no official recommendation is mentioned.

I found this Docker network_mode comparison table on LoadForge:

https://loadforge.com/guides/advanced-network-configuration-for-high-performance-docker-containers#:~:text=The%20host%20network%20driver%20provides,network%20throughput%20and%20low%20latency

Has anyone already benchmarked these differences specifically for Jellyfin workloads?  When direct-streaming large 4K HDR remuxes (65Mbps) over the internet (~20 miles), I frequently get micro-pauses/suttering. Could switching to a more secure/isolated mode like bridge worsen this?

Thanks,

P.S. I’m already considering pre-transcoding static HDR titles later down the line.

[TheDreadPirate]

In terms of performance and connection stability, the only issues I've seen from other users is when paired with a reverse proxy that is also in a container.  When your reverse proxy is also using bridge networking, you should not also publish port 8096 for Jellyfin.  This results in excessive "context switching" between the bridge network and the host network.  Not sure if that is that is the right term.

   

But this is OK.

   

If your reverse proxy is NOT a container, but runs directly on the host, the second image would result in pretty much the same thing.  But would require you use "expose" for port 8096, which is similar to port forwarding on your router.

Code:

    expose:
      - 8096

The idea is to reduce how many hops in and out of the bridge network your traffic makes.

But, I've only seen the first image cause connection stability issues for some users.  I don't think there is a performance issue.  No additional CPU overhead or limited throughput.