[Bioclusters] Large numbers of files (again...?)

Andrew Fant bioclusters@bioinformatics.org
Tue, 03 Feb 2004 10:12:02 -0500

--On Wednesday, January 28, 2004 15:59:52 +0000 Dan Bolser 
<dmb@mrc-dunn.cam.ac.uk> wrote:

> ++ Joe Landman--
>> Hi Dan:
>>   If it is a chore to have them switch, then the hashing function is a
>> much better choice.  I am not so sure about trusting real data to
>> Reiser, I have been (badly) burned by it in the past.
> I know Reiser and xfs are still very new.
> Anyone actively using xfs?
> I guess switch involves reformatting disks? Actually we have a large
> storage space yet to come 'on line'. It might be a good chance to try xfs.

Actually, xfs is new only in the sense that it has just made it into the 
stable linux kernel trees.   As an old SGI admin, I've used it for about 5 
years total, and about half that long on Linux.  As you might suspect, I am 
quite fond of it.

On my Linux systems, I use it for most partitions except for /boot (grub 
can handle xfs, but with lilo and multiple os use, xfs on a boot partition 
is a no-no).   A couple caveats apply.  First, you should always have your 
system on a ups of some kind if you use xfs.  The aggressive caching is 
great, but a badly timed power cut can trash a filesystem ( It's never 
happened to me, but forewarned is forearmed).  Also, to get the best 
performance, you need to do a little reading of the documentation for 
mkfs.xfs.  The defaults tend to over-allocate internal filesystem data 
structures, which can slow things down.

If there's anything more specific you want to know, please feel free to 
drop me a line.