On Wednesday 26 April 2006 10:30, McGraw, Robert P. wrote:
> 1) The above log seems to show that the tape_splitsize is 10G and not the
> 30G that I indicated in the tape_splitsize. Is this correct? Other than the
> tape_splitsize what other parameter do I need.

No, the tape_splitsize used was only 10M, not even 10G. Read on.

> 2) My backup was running slow and and noticed that
> zorn./export/fssnap/users was not in the HOLDING DISK so it must be reading
> the backup directly from the disk. My holding disk is 200GB and the
> filesize that I am backing up is 300GB. Do I need to have my HOLDING DISK
> size as large as the file system that I am backing up when using the
> tape_splitsize?

Yes, the spanning and splitting functionality still does not allow you to 
interleave dumps from different disks. So, if you do not have a big enough 
holding disk, dumps will still be done directly to tape -- which can be very 
slow.

When dumping directly to disk, and mmap is not available, then we use the 
fallback_splitsize (default 10M) instead of the tape_splitsize -- this is 
because the buffering is done in memory. If you have the memory, feel free to 
set a larger fallback_splitsize -- it will improve performance a lot.

Another option is to break the disk into smaller chunks that will fit on the 
holding disk. See here:
http://wiki.zmanda.com/index.php/Manually_splitting_large_disks

Cheers,

--Ian
-- 
Forums for Amanda discussion: http://forums.zmanda.com/