"NFS protip: if you remote-mount your NFS filesystems, soft, with retries, retransmits, and maybe timeouts set to appropriately high values, you can create a window of opportunity for maintenance, of arbitrary width - we recommend 72 hours, that's usually enough time to have hardware fail, call your technical support representative, arrange for the new part to be airshipped overnight, meet the field engineer the next day, install the new part, and still have 24 hours left to handle any software problems that emerge - where your NFS clients won't lock up when an NFS server goes offline AND you will not lose any data from pending transactions from your NFS clients, just as if it were an NFS hard mount.

We developed this methodology - in conjunction with our coworker, Dr Peter Schachte - nearly forty years ago, at Quintus.

NFS-based supercomputer cluster administrators, take note; all of the UNIX/Linux clustering software we have worked with mounts NFS filesystems hard, and we predict that your entire NFS cluster and all its NFS clients lock up every time you reboot one of your NFS servers. With this one small change you might eliminate 75% of your downtime, or more."

Please discuss.

Guru move?

Obsolete?

Just plain wrong?

Teach me.