[NTLUG:Discuss] [Bulk] Re: ext3 waste disk spaces then Windows ME?

[Chris Cox]

Robert Pearson wrote:
> On 4/30/06, David Stanaway <david at stanaway.net> wrote:
>> Chris Cox wrote:
>>> At the time I did my work comparing them, XFS wasn't quite baked yet
>>> (had some issues).  XFS has always been the filesystem to wait for...
>>> and now that it is here, perhaps it is an excellent choice.
>>>
>>> Btw, my findings agree with the findings in the article with regards
>>> to the low CPU utilization characteristics of JFS.  We went with
>>> reiserfs because it was it was very fast on some operations and not
>>> terribly far off the mark on others, and what we really liked was
>>> the ability to resize filesystems on the fly (while still mounted).
>>> Which at the time we did our tests, only reiserfs could do this.
>> xfs does that too. It is a great compliment to lvm2.
> 
> Just an FYI...
> According to both "LVM HOWTo and the "OSS SGI Project"
> LVM cannot shrink an XFS Filesystem.
> "LVM HOWTO" says LVM cannot shrink a JFS Filesystem.
> 
> <<http://www.tldp.org/HOWTO/LVM-HOWTO/reducelv.html>>
> 
> <<http://oss.sgi.com/projects/xfs/faq.html#resize>>
> 
> These may be important considerations for an environment.
> They can all be expanded.
> 
>> One more serious that had occurred with xfs in the past was a bug with
>> stale block data after the end of the used portion being seen when a
>> file was read with memory mapped IO as occurs with gcc for instance.
> 
> The OSS SGI site reports some bug fixes---
> 
> <<http://oss.sgi.com/projects/xfs/index.html>>
> 
> Notice the copyright expires in, or at the end of, 2006.
> 
> While I love XFS and believe it is a fine filesystem, it is the
> fastest one I ever benchmarked, I would wonder about
> future SGI support.
> SGI needs XFS for their HPC boxes they sell the Government,
> and Big Oil, but they are a company living on borrowed time.
> If Red Hat had not picked up GFS one of the best filesystems
> for Storage Management would have disappeared. GFS is the
> only filesystem I ever saw that was Storage Centric and Storage
> Aware. I have given up waiting for those features to appear in
> any other filesystem.
> There is a case to be made for "the way things are".
> Give people the fastest filesystem you can and let them make
> it do what they want. XFS is fun to customize.
> 
> IMHO
> These considerations only apply at the Enterprise. With some
> trickle-down to SMB environments. At the Personal Computing,
> SOHO and low-end SMB you are free to be creative.

You know there is a need for filesystems that cater to specific
needs.  For example, a large multi-terabyte filesystem will
take more than a day to fsck.  Journaling reduces the need
for an fsck, but occasionally it is necessary.  Who wants to
have leave a filesystem offline for over a day??

Another issue it the time it takes to delete a tree of data
with thousands and thousands of files.  It can take FOREVER!!

None of the filesystems out there deal well with these two
cases today (AFAIK).

XFS, jfs, ext3 or reiserfs, they all have advantages and
disadvantages.. but I know, I'm still waiting for yet one
more filesystem.  One that takes into account the large
storage needs that are becoming popular now.

And I do like GFS and look forward to its universal
availability (it's not quite there yet).