Hi,
(Basicaly you have to get RDMA working first. Look here: https://enterprise-support.nvidia.com/s/article/howto-configure-nfs-over-rdma--roce-x)
it took me a while to find a way how proxmox could use NFSoRDMA. First I tried to change the storage options in /etc/pve/storage.cfg from
	
	
	
		
to "options rdma,port=20049,vers=4.2" but it was still mounted with TCP. (Why?)
But there is a way, if you mount the NFS share to some other folder with RDMA option before the pvedaemon mounts the storage, it will also get mounted with RDMA instead of TCP as protocol.
	
	
	
		
	
	
	
		
@ ProxmoxDevs
It would be nice if this could be done more elegant in future.
				
			(Basicaly you have to get RDMA working first. Look here: https://enterprise-support.nvidia.com/s/article/howto-configure-nfs-over-rdma--roce-x)
it took me a while to find a way how proxmox could use NFSoRDMA. First I tried to change the storage options in /etc/pve/storage.cfg from
		Code:
	
	nfs: NFSoRDMA
        export /testdir
        path /mnt/pve/NFSoRDMA
        server 192.168.40.10
        content images
        options vers=4.2
        prune-backups keep-all=1to "options rdma,port=20049,vers=4.2" but it was still mounted with TCP. (Why?)
But there is a way, if you mount the NFS share to some other folder with RDMA option before the pvedaemon mounts the storage, it will also get mounted with RDMA instead of TCP as protocol.
		Code:
	
	mount -o rdma,port=20049,vers=4.2 192.168.40.10:/testdir /testdir
		Code:
	
	#mounts
192.168.40.10:/testdir on /testdir type nfs4 (rw,relatime,vers=4.2,rsize=1048576,wsize=1048576,namlen=255,hard,proto=rdma,port=20049,timeo=600,retrans=2,sec=sys,clientaddr=192.168.40.11,local_lock=none,addr=192.168.40.10)
192.168.40.10:/testdir on /mnt/pve/NFSoRDMA type nfs4 (rw,relatime,vers=4.2,rsize=1048576,wsize=1048576,namlen=255,hard,proto=rdma,port=20049,timeo=600,retrans=2,sec=sys,clientaddr=192.168.40.11,local_lock=none,addr=192.168.40.10)@ ProxmoxDevs
It would be nice if this could be done more elegant in future.
			
				Last edited: 
				
		
	
										
										
											
	
										
									
								 
	 
	