Link root-painter to a working HPC enviroment running slurm

[paubach711]

Hi, I installed root-painter in one of my HPC servers running slurm, its seems to work but i dont really know how can i link the root-painter trainer to the server.
Usually the user im helping runs this software with google collab but we want to run this on more powerful machines and locally

[Abe404]

Hi,

I'm not sure exactly what you mean by 'link root-painter trainer to the server'. Could you provide a little more detail?

The way RootPainter works is there is a client and a server component. They communicate via a shared folder. If your slurm nodes have access to a shared, synced or mounted drive that the user can also mount/sync on their desktop / workstation where they run the client, then it should be possible to connect the client and server. The key is you have the root_painter_sync directory specified inside this shared drive/folder for both client and server. The server asks for you to specify a root_painter_sync directory when you first start it (it is subsequently stored in a root_painter_settings.json file in your home folder. The client asks for you to specify this directory when you first run it (and it is also subsequently stored in a root_painter_settings.json file in your home folder). Updating or locating the sync directory with the client GUI is also possible via options in the extras menu.

I usually use sshfs for mounting drives on a server, but perhaps your university already has some type of shared drive/folder system that would work for this particular user?

With colab the shared folder is facilitated with google drive, but I suggest finding a better solution for your own HPC setup as google drive is slow to sync data and can cause data sharing concerns for many universities (depending on their policy).

Another challenge with slurm and why I don't personally use it, is that it often assumes jobs can be run in the 'background' at some time in the future. RootPainter facilitates a real-time interactive process for model training that often doesn't fit the way some universities (or other organisations) like to have their slurm jobs scheduled.

Kind regards,
Abraham

[Abe404]

FYI I'm converting this issue to a discussion as it's not clear to me that there is an issue here.