On Tue, 2 Mar 2004 at 3:48pm, Jonathan Dill wrote

> On another note, maybe things have changed, but I once found that gnutar 
> incremental backups sucked performance-wise, would make machines pretty 
> much unusable during estimates and dumps.  Normally, this would not 
> matter, but you're talking University with eccentric grad students 
> working at 3am and such who complain about these things.  I have 
> migrated most things to XFS filesystem and use xfsdump on Linux and 
> IRIX--a process that I started when XFS went Open Source (around Red Hat 
> 7.0) and I got tired of waiting for the problems with dump for ext2fs to 
> get sorted out.  Machines are still very usable with xfsdump and 
> software compression running in the background, and finish faster than 
> gnutar dumps.  xfsdump estimates are very fast, comparatively speaking.

XFS and xfsdump are indeed very nice.  But filesystems like this:

[jlb@$HOST jlb]$ df -h
Filesystem            Size  Used Avail Use% Mounted on
.
.
.
$SERVER0:/data          535G  518G   18G  97% /data
$SERVER1:/moredata      1.8T  1.2T  621G  66% /moredata
$SERVER2:/emfd          2.0T  779G  1.3T  39% /emfd

make tar rather necessary (those are all XFS on Linux servers BTW).  For 
the record, estimates on those servers go *very* fast (<5 min).  I *do* 
have one server with a 1T XFS filesystem that takes a *long* time to 
estimate one particular direcotory (~90 minutes).  But I'm pretty sure 
that's due to an inordinately large number of tiny files and 
subdirectories in there (about which I'm beating up the user).

-- 
Joshua Baker-LePain
Department of Biomedical Engineering
Duke University