[NBLUG/talk] Recommendations for journalling filesystems?

[Lincoln Peters]

On Jul 27, 2006, at 9:42 AM, Eric Eisenhart wrote:

> I recommend sticking with ext3 unless you have specific needs that it
> doesn't meet.  It's very mature with the benefit that it can be  
> treated as
> if it's an ext2 filesystem.  With some specific needs you might  
> just need to
> tweak some of the ext3 options such as dir_index (speed up lookups  
> in large
> directories with a btree), block to inode ratio, journalling data  
> instead of
> just metadata, non-ordered journalling and various block-size options.
>
> For instance, if you wanted to run a big, high-performance NNTP  
> server with
> a terabyte or three of ~10K files with some directories containing a
> hundred thousand files, ext3 wouldn't be the best choice.

I don't quite deal with hundreds of thousands of files in a single  
directory, but I do sometimes deal with directories containing a few  
hundred files.  And I do have a few directories with some large  
files, although I don't have anything on the scale of MythTV video.

>
> One thing to remember is that XFS and JFS were originally designed  
> with the
> needs of enterprises in mind.  Places where SCSI (with automatic  
> badblock
> handling built into the HD firmware), RAID, daily backups to tape, and
> notifications of potential drive problems before they're really a  
> problem
> are the norm.  In other words, they might anticipate that the way  
> to handle
> the (very unusual in the above situation) problem of a totally  
> failed drive
> (errr, failed RAID container) is to swap in new drives, recreate  
> the RAID
> group and restore from backup.

Let's see...

Automatic badblock handling:	I think that IDE hard drives do that  
automatically now, at least to some extent.

RAID:	I'm already using it--one RAID-1 array, and one RAID-5 array.

Daily backups to tape:	Once all is said is done, I'll be using  
something equivalent (I can do daily backups to the external hard  
disk I just ordered).

Notifications of potential problems:	Isn't this more or less what  
S.M.A.R.T. does?  I think I can use smartd to do periodic checks on  
all the hard drives.

Replace totally failed hard drives:	Given a working RAID system, I  
think I can do this.  Especially if I'm using Seagate hard drives  
that come with a 5-year warranty.

Recreate the RAID groups:	Easy when using mdadm, although it does  
sometimes entail a significant performance penalty while it's  
rebuilding.

Restore from backup: Should be easy and very fast with an external  
hard drive.


Suddenly these assumptions don't seem so far out of reach now.  And  
if Debian supports XFS, that might just make XFS a viable option...


--
Lincoln "The DiskBuster" Peters
<petersl at sonoma.edu>

Computer programmers do it byte by byte.