On Oct 29, 2010, at 7:43 AM, James A. Peltier wrote: > As for SFTP or any other method that would duplicate data, I have already discussed why it is not a possibility. SSHFS *was and still is* a possibility but it was ruled out because of our HPC needs.
I run something that could be considered (and is often referred to as) an HPC cluster. We leverage NFS heavily. We'd melt our filers if they weren't front-ended by NFS caches. You can't seek() using sftp. You can't lock a file using sftp. It's a bitch to code in sftp support to every application that expects to operate on a file. And it scales for shit: Suck down a file and wait for the whole thing? Run sshfs on the cluster to a centralized sshfs-based fileserver? I don't think so. We are using NFSv3. We'd love to have delegations in NFSv4 because it would significantly enhance the ability to locality-based locking/caching. Thousands of machines sharing the same multi-petabyte dataset won't work with sftp. Or sshfs. My point is not to suggest what the OpenBSD developers should or should not implement. That is there decision. But it annoys me that people think sftp (or any other non-block-based file transfer mechanism) is a replacement for NFS. It's not. And it's not to suggest NFSv4 is the bees knees. Some people may need/want it. Some may not. Sean PS I can't believe I got sucked into this thread.
