RFC: Tuning ext3

[Mingming Cao]

On Thu, 2007-05-17 at 18:25 -0400, Tod Hagan wrote:
> On Thu, 2007-05-17 at 23:25 +0200, Ralf Hildebrandt wrote:
> > * Andreas Dilger <adilger at clusterfs.com>:
> > 
> > > > E. Use a kernel >= 2.6.19 (patches for extents and 48-bit support,
> > > > requires Ubuntu 7.04 feisty or Fedora Core 7 or custom kernel) to allow
> > > > filesystems > 8TB on Intel/AMD chips. [BigFS]
> > > 
> > > This means ext4.  I'd suggest including the mballoc and delalloc patches
> > > from Alex if performance is the driving factor.  There is some risk
> > > involved in ext4.
> > 
> > I don't understand this: The whole idea was to stay "supported", and
> > now the OP want's to use hand-built kernels? Why?
> 
> Yes, this item contradicts the constraint of using RHEL-supported ext3
> and will be removed from the next revision of the document. In its place
> will be a note that ext3 filesystems > 8TB on Intel/AMD chips aren't
> possible on vanilla RHEL.
> 
Do you mean >8TB or >16TB?

Linux kernel >2.6.18 fixed ext3 kernel bugs so ext3 can support >8TB
filesystem. The filesystem size is still limited up to 16TB due to 32
bit block numbers (based on 4k default block size). It needs
e2fsprogs-1.39 to able to create a fs >8TB.

To get >16TB filesystem you either move to ext4(avaible >= 2.6.19) or
apply 48bit extents patch though, the later is not in mainline. 

Mingming