Provide OVF support
bigfoot1942 opened this issue · 9 comments
If you were to offer an OVF containing a virtual appliance that can use nVIDIA Tesla T4 videocards, using vGPU, that would greatly reduce effort to implement this in an enterprise environment.
I have only tangential experience with OVF/OVA files, but I am happy to help. I have some questions though:
- Wouldn't we have license issues if we repackage things such as the Nvidia drivers or CUDA binaries?
- What's the expected OS for the OVA? I guess a Linux flavor?
- Wouldn't Docker images serve the same purpose here? Is there any chance you can use those to help you create the OVF/OVA? There's also Azure templates.
Ok, I have tried my best to export amalic's Docker image to an OVF/OVA appliance. You can get it here: https://we.tl/t-fn2IoAt8LZ
I followed these steps:
- Cloned https://github.com/iximiuz/docker-to-linux and edited the Ubuntu recipe so it uses amalic's docker image instead.
- Added
vi
,nano
andnetwork-manager
to the apt install instructions. make ubuntu
qemu-img convert -i raw -O vmdk linux.img fah.vmdk
- Created a new VirtualBox VM using that VMDK file. At this point you can login with
root:root
. - Patched
/etc/fstab
to avoid read-only bugs. - Patched
/etc/init.d/FAHClient
so it uses theroot
user (otherwise it fails to start). - Exported the whole thing as an OVA appliance.
Basically, FAHClient will start in daemon mode even if you don't login, so that's all set up. You might want to edit /etc/fahclient/config.xml
to specify username/team and/or enable more slots.
BUT:
- I don't have access to VMWare to test it.
- I haven't tried to expose the GPU to the VM.
- I don't have vGPU-capable hardware so I can't test it either.
Hopefully this is enough to get you started?
I am not aware of any more efforts being done, but given the popularity of F@H at this moment, I wouldn't be surprised! @seanpmassey, can you offer some more details here? Thanks!
Awesome, thanks for the report!
For those wondering, OVA support is already here: