Different kernelheaders for gpu use? do i need seperated boot images? can i share the hardware-specific software?

LINSTOR
1

Hey im pretty new to drdb and im currently looking for a way to provision my servers in a diskless environment where i can choose if a node should be used to run a vm, container, or be k8s node. So far i have learned foreman and set it up with external dhcp & dns in a proxmox vm. i know how to set up a diskless scenario where i can choose images for pxe boot based on parameters or hostname, so that shouldnt be a problem. but i still dont know if drdb will provide what im looking for.

so is it theoretically possible to boot the proxmox node with the desired kernel headers and share the workloads among this group that include the drivers, cuda/rocm and stuff like container toolkit without installing it on every node?
so im actually looking for a better alternative for booting several servers from the same drive.
best reagrds: ji
and greetings from germany

2

I couldn’t get any answers to this question, even through AI, Discord or Reddit, and this problem has been bothering me for more than a year, so I’m trying this approach and hope it finally solves my headache :slight_smile: I’m wondering how that is done in data centers

3

So you want to share a “root” disk over the network, right? While theoretically possible with DRBD, it’s not really designed for this. For starters, you would need to reconfigure the “storage node” every time a new server connects to expect the connection.

What you likely want is hosting the root disk using iSCSI or NVMeoF.