Beiträge von midisama

Zitat von ryecoaaron

How did you envision entering that info in a GUI? It isn't really possible to make the web interface dynamic based on number of cpu cores. I am curious why you need this though. I have 20+ VMs running on an e5-2697v3 and don't have stuttering problems without this.

Either through the Change vCPU menu, you can list the thread siblings of each core (that can be listed through this command: cat /sys/devices/system/cpu/*/topology/thread_siblings_list | sort -nu) and in the VM creation page if the advanced section is enabled. It can be shown as a list with threads and check marks:

CPU0: Thread 0 [ ] - Thread 14 [ ]
CPU1: Thread 1 [ ] - Thread 15 [ ]
...
CPU12: Thread 12 [*] - Thread 26 [*]
CPU13: Thread 13 [*] - Thread 27 [*]

Isolating CPUs really helps with gaming performances in a VM, I have 2 nvidia GPUs each set for a different VM, if I do not isolate the cpus as before, the frametimes jump to the roof, by isolating the cores/threads, it removes any noise from other processes which leaves them exclusively for the vm (or docker) and leads to better performances.

For the docker section, it's just an option that you can add for select CPU threads (only works with non-isolated CPUs, meaning they're enabled and shared with the host). I do not have any real world use for it, but some people might need it, so I suggested it (ofc, you have the final word in that, you can also just set a "CPU Set" section where people enter their own thread counts like "1-3,6-8" or the like, it's per container/service basis).

Hi,

I am currently using cpu isolation and pinning to lower stuttering in my vms. For now I'm using terminal and manual xml editing for that, I set `isolcpus=5-13,19-27` in grub and set cpu selection manually in the xml like the following:

  <vcpu placement='static'>8</vcpu>
  <cputune>
    <vcpupin vcpu='0' cpuset='5'/>
    <vcpupin vcpu='1' cpuset='19'/>
    <vcpupin vcpu='2' cpuset='6'/>
    <vcpupin vcpu='3' cpuset='20'/>
    <vcpupin vcpu='4' cpuset='7'/>
    <vcpupin vcpu='5' cpuset='21'/>
    <vcpupin vcpu='6' cpuset='8'/>
    <vcpupin vcpu='7' cpuset='22'/>
    <emulatorpin cpuset='1,15'/>
  </cputune>

I would like to see this possible in the gui directly if possible. Systemd can handle the cpu isolation instead of using the command line (example: https://www.reddit.com/r/VFIO/…cpu_pinning_explaination/).

Thanks.

Zitat von ryecoaaron

It isn't adding the options because you aren't passing a gateway. I guess I will have to change the plugin to allow this.

you mean "IP Range"? Because in my screenshot I do have a gateway. I usually setup my IPs manually in the compose file, I do not want them to be dynamically allocated.

Hi,

I saw that the network section has now Subnet, Gateway and IP range sections for macvlan, but it doesnt seem to be using the values that I put in it, here are my values:

but here is what I get from the inspect:

[
  {
    "Name": "netbr",
    "Id": "df46b4822e54ed49b09596e15a3436526232b381595772a96be4d32500c2be41",
    "Created": "2023-07-10T15:11:39.375777423+01:00",
    "Scope": "local",
    "Driver": "macvlan",
    "EnableIPv6": false,
    "IPAM": {
      "Driver": "default",
      "Options": {},
      "Config": [
        {
          "Subnet": "172.19.0.0/16",
          "Gateway": "172.19.0.1"
        }
      ]
    },
    "Internal": false,
    "Attachable": false,
    "Ingress": false,
    "ConfigFrom": {
      "Network": ""
    },
    "ConfigOnly": false,
    "Containers": {},
    "Options": {
      "parent": "br0"
    },
    "Labels": {}
  }
]

END OF LINE
█

I use this command to do the same thing:

docker network create -d macvlan \
    --subnet=192.168.0.0/24 \
    --gateway=192.168.0.1  \
    -o parent=br0 \
     netbr

which works. This might be a bug in the plugin.

Zitat von Cap10Canuck

Any luck? I am in the same boat, and the only advice I got I think was pointing me to your post! My nividia integration is completely messed up since the change. I managed to get all my other stacks up and running relatively easy, as long as I commented out ANY reference to nvidia. However, I need to include references to nvidia in both Plex and Jellyfin stacks for them to work, otherwise I completely lose transcoding ability on my server. Any pointers appreciated.

It works! as long as the path in the gui is empty, the json retains whatever's in it .

Zitat von ryecoaaron

You leave the path blank in the settings tab if you need to customize daemon.json.

Alright, I will test that.

Hi, I'm experiencing an issue with keeping nvidia runtime properly set in `/etc/docker/daemon.json`. Everytime I apply any change in the Compose gui in OMV, the daemon.json file gets reset and removes nvidia runtime section.

Here is how I set it up

{
  "data-root": "/DOCKER/system/",
  "runtimes": {
        "nvidia": {
            "path": "nvidia-container-runtime",
            "runtimeArgs": []
        }
    }
}

But it gets reset whenever I apply changes.