Dear all,
Regardless of the recent "flame wars", I, nevertheless, need to
ask for your learned opinion on tar, Amanda and Linux. We have been
performing archive backups with GNU tar on a Linux box past Amanda and
made the following observations: The default blocking factor of 20
performs really poorly over network. Increasing the blocking factor
from 20 to 126 improves performance tenfold.
We are now in the process of redefining our amanda configuration to use tar
because of the aforementioned drawbacks of 2.4.X kernels, ext2 and dump.
We are willing to accept the changed atime/ctime over the imminent data
corruption, as the clients volumes are all ext2.
My question is this. What blocking factor does Amanda use (with tar)? I
have a vague recollection that the blocking factor is 32 1K blocks,
which is 64 POSIX blocks. Is there any way to tweak the blocking factor?
I guess the "runtar" binary is the one that feeds the backup stream to
dumper? I wasn't able to find any documentation on it, and a quick dive
to the client-src directory didn't yield any info on the arguments
runtar is called with:
runtar.c: execve(GNUTAR, argv, safe_env());
"sendbackup-gnutar.c" on the other defines the options, but there was no
reference to the blocking factor.
Furthermore, how dependent on the blocking factor is Amanda? Can it be
configured to use 63KB blocks, which in terms of network performance
has proven to be the most efficient and portable way to run GNU TAR?
(I have been lead to believe that using 64 1K blocks may wrap around on
certain platforms.) In terms of command line switches it might be
beneficial to use "--atime-preserve" on certain volumes and disable
it on others. Is there any way to do that?
Please feel free to correct any misconceptions I may have voiced :-)
Regards,
Lari Huttunen
--
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
| Lari Huttunen | +358 50 338 0759 | http://www.helsinki.fi/~lshuttun |
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
