From: utzoo!decvax!harpo!floyd!vax135!ariel!houti!hogpc!houxm!ihnp4!ixn5c!inuxc!pur-ee!uiucdcs!parsec!kolstad
Newsgroups: net.unix-wizards
Title: Re: Thoughts on system backups - (nf)
Article-I.D.: uiucdcs.1843
Posted: Thu Apr  7 23:36:14 1983
Received: Tue Apr 12 23:07:50 1983

#R:ittvax:-67000:parsec:44200004:000:1433
parsec!kolstad    Apr  7 11:08:00 1983

Restoring directory trees is an awful experience whether it is
labeled 'trivial' or not.  We recently had a disk drive failure and
wished to restore two dozen users from /mnt to a file system on our
remaining disk.  The construction of the restor/mv files was awful
(mkdir's have to be included also; our shell accepted only 10000 chars
for single commands).

I disagree that using the tool called 'restor' is the correct solution.  I have
modified (a much nicer word than hacked) restor.c to create retrieve.c:
it (a) allows a trailing * on file/directory names and (b) creates
directories automatically in the context of (a).  This allows trivial
restores of tree structures (with only a single pass across the multiple
tapes).

The program is currently distributed to four sites.  It has a performance
problem during restore since it has to check the inode number from the tape
against EVERY possible retrieved inode number -- currently by using a
linear search.  The program has NOT been thoroughly tested with bizarre
conditions.  In a week or two, I'll submit it to net.sources and prepare
to suffer through the incredible amount of comments about lousy user
interface and poor implementation strategy (that are ALREADY a part of
restor.c).  

It is unfortunate but true that the new retrieve.c program uses a large
list that can be supported only on virtual unix systems.

				Rob Kolstad
				PARSEC Scientific Computer Corp.