At our lab we generate fluid flow simulation data which is done overnight and we take the data for other analysis purposes. Its about 30TB and we would like to backup the data to another lustre filesystem. In addition to backup I would also also to get high availability. So, I was thinking of this on a client /data (NFS mountpoint) /fd0 Lustre, Fluid Dynamics filesystem. Coming from mgs "prod" /fd1 Lustre, Fluid Dynamics filesystem. Backup, coming from mgs "bkup" /data/fd0 is a symbolic link from /fd0 The data is replicated from /fd0 to /fd1 every morning to keep it in sync. Using rsync If /fd0 is every down, we just change the symbolic link from /fd0 to fd1 and get fd0 up ASAP. Is this a good strategy? I kind of want to avoid drbd because of the write penalty and rsync is pretty simple. I was wondering if anyone else has some better techniques. Also, is there anything I would need to do on the clients to avoid the blocking? I don''t want to the client to keep waiting for /fd0 to comeup.