Jim Rees (email@example.com)
11 Aug 87 19:34:00 GMT
Sun's ND (one of the weirdest beasts around) isn't stateless, you
can't use that as an example in this case.
I don't think he was talking ND. NFS clients will get confused if the
server's handles change, which they will if you mkfs and restore.
The handles are some combination of inode number, device number, and
Some of this confusion could be lessened if you could guarantee that
handles will never get re-used. This is the approach we took in our
NFS server. Instead of inode numbers, objects are named by UID. The
UID (Unique ID) is a combination of node number and timestamp, and is
guaranteed unique, even if you mkfs and restore. We didn't invent the
idea, it's a commonly used technique in more modern (than Unix) operating
systems. With UIDs for handles, the clients at least don't mistake
new objects for old ones, although they still can't find the ones they
You could argue that this is correct behavior anyway, since when the
files come back in off of tape, they probably aren't the same as they
were immediately before the crash.
This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:38:49 GMT