I think that part of what planner uses to decide which image to tell taper
to write next is based on what images are available in the holding disk to
write to tape.
If you want to affect that order, there are ways (in the amanda-users
archives) to delay the start of individual dumps. You could delay the
smaller dumps until the larger ones should be complete. (Or at least writing
to tape.)
Alternatively, if you are always going to two tapes anyways, you could run
two separate configurations, and balance the large backups across them by
hand, so that each one always fits on one tape. (Once you divide them,
AMANDA will help pick backup levels to fit the tape.)
Also, there was a discussion on amanda-hackers about a 'chunking' method
that would be able to split an image across tapes, among other neat things.
It's in discussion for a future version of AMANDA. Parhaps you could
contribute to that effort as a coder and/or tester?
-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]
Sent: Wednesday, May 02, 2001 3:08 PM
To: [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]; [EMAIL PROTECTED]
Subject: Re: How amanda thinks
I've thought that the reverse behavior (largest images first) would
result in better usage of tapes in a multiple tape configuration. Our
offsite backups run to multiple tapes, and Amanda always finds the end
of a tape by writing off the end of it:
INFO taper tape offsite-009 kb 19799936 fm 56 writing file: No space left
on device
which results in the large image being written to the next tape.
Now, perhaps my Amanda configuration is screwed up, but if so I don't
know how. I'm writing to DDS-4 tapes, with a tapetype definition of
19,500 mbytes and a filemark of 0 bytes. My most recent log has
17,863,648 KB being written to the first tape, and then trying to
write an additional 5,899,168 KB, which would result in putting a
whopping 23,762,816 KB on the tape, far in excess of the tapetype
definition.
It struck me that doing the largest backups at the front of the tape
would make this work better.
But perhaps I have something messed up someplace. :)
