| 1 |
On Wed, Apr 11, 2007 at 02:06:44AM -0700, Wil Reichert wrote: |
| 2 |
> On 4/10/07, Joshua Hoblitt <jhoblitt@××××××××××.edu> wrote: |
| 3 |
> >After looking through various patches that were posted to LKML, I |
| 4 |
> >discovered that passing "-F" to mke2fs seems to make this work as |
| 5 |
> >expected. However, this behavior seems to be undocumented and that |
| 6 |
> >makes me more then a little nervous. |
| 7 |
> > |
| 8 |
> >The resulting filesystem does seem to be mountable: |
| 9 |
> > |
| 10 |
> > /dev/sda4 11T 173M 11T 1% /export/ipp003.0 |
| 11 |
> > |
| 12 |
> >Is this feature undocumented for a reason or is it just unsupported? |
| 13 |
> |
| 14 |
> Any reason you are so set on ext3? Its fine for a general purpose fs |
| 15 |
> but this might be a sign you are stretching its intended limits. |
| 16 |
> Perhaps something 64-bit like xfs or jfs would be a better option |
| 17 |
> depending on your intended usage / workload. |
| 18 |
|
| 19 |
Yes, ext3 has proven itself, to my satisfaction, to be extremely |
| 20 |
reliable even if it's not mounted with the journal in ordered mode. In |
| 21 |
fact, I've never lost data from an ext3 filesystem that wasn't the fault |
| 22 |
of the hardware (disk/controller failure). I've dabbled with xfs and |
| 23 |
had some issues -- I've spoken with people that have had data corrupt |
| 24 |
issues with xfs when the system has exhausted memory. For my usage |
| 25 |
pattern, xfs only gives me about a 10% performance boost and frankly I |
| 26 |
don't need the extra bandwidth. I don't really have any experience with |
| 27 |
jfs. At least to me, ext3 is clearly the most stable choice. Even |
| 28 |
lustrefs uses ext3 as the storage backend because of it's stability. |
| 29 |
|
| 30 |
-J |
| 31 |
|
| 32 |
-- |
Archives