Depends on what you are using, for connecting block stores, you'll use some sort of fabric. That is Fibre channel, SAS, NVME over something or other If you are

Depends on what you are using, for connecting block stores, you'll use some sort of fabric. That is Fibre channel, SAS, NVME over something or other If you are using GPFS, then you can do stuff over IB, but I don't know how that works. Lustre I imagine does lustre things over RDMA. For everything else, NFS all the things. pNFS means that you can just throw servers at the problem and let the network figure it out. But again, if IO speed is critical, you move IO over to a dedicated fabric of somesort. for most thing NFS is good enough. (except databases, its possible but not great. but then depending on your docker setup, you might be kneecaping your performance because overlayfs is causing io amplification)