> We're looking at some tape stackers and the possibility
> of setting something like expect up with them to automate
> backups that run into the wee hours.
>
> Has anyone added stackers or silos or such to their afs
> backup procedures? If you have, I'd be really interested in
> knowing what hardware and software combinations you're using.
> In particular, I'm interested in whether you were able to
> wrap an expect script or something similar around the butc
> processes and thereby achieve cheap(er?) set up time on the
> stackers? Or did you have to (or choose to) do major brain
> surgery on the backup sources?
I did not use 'expect', my setup is primitive enough such that a standard Korn
shell wrap around butc suffices which essentially handles 'hitting the return
key' as soon as the butc message starts with the magic keyword 'Please'.
Backups go to an Exabyte 8500 in a 10 cartridge stacker in 'roll' mode.
The backup script starts butc, reads the tape label and performs a dump
as indicated by the label. This means that the tape decides which dump
to start. The dump sets are dimensioned such that they fit on one tape.
At the end of the dump, the backup script ejects the cartridge and the stacker
fetches the next one. This goes on 'forever' unattended except that somebody
has to replace worn out tapes from time to time.
You can see from this that we have yet relatively little data to backup.
I have at least the following worries as the file base grows:
1. IMHO Transarc's technique to rely on a fixed capacity in tapeconfig has no
future. Running the 'fms' utility to get the figures for the 8500 already takes
about a day. And what about data compression? The only solution seems being
able to write until end-of-tape.
2. The butc interface is insufficient and in fact unsuited for programming
around it with utilities like 'expect': if you were to handle the request for a
follow-on tape you'd have to be able to label that tape first, currently
impossible since a job requesting that tape is already running. Also, the
'expect' technique has it's limits. I'd rather see butc call a set of
customizable shell scripts which in turn can use a limited set of backup's
functions like readlabel and labeltape.
3. backup/butc decide on the tape label, but I wonder how I could uniquely
relate that one to external labels or directory information for a robot?
4. User-initiated restore in my setup is obviously unimaginable.
--
Rainer Toebbicke - [EMAIL PROTECTED] - [EMAIL PROTECTED]
European Laboratory for Particle Physics - CERN
Geneva, Switzerland
