On Thu, 2002-04-25 at 05:09, ASANO Masahiro wrote:
> Thank you for your reply.
>
> From: Steve Lord <lord@xxxxxxx>
>
> > In order to deal with the 32 bit limit on Linux, and with issues with
> > 3rd party applications such as Legato Networker on Irix, changes
> > went in recently which adjust the inode placement policy of XFS to
> > restrict inodes to the part of the filesystem which keeps them within
> > 32 bits, unless the inode64 mount option is used. The fix is more
> > complex than that as we need to make data prefer space not suitable
> > for inodes to avoid filesystems which have lots of space but cannot
> > create new files.
>
> But it seems that XFSMNT_32BITINODES flag is always set.
On Linux yes, since the upper layers of the vfs impose this limitation
on the size of the inode. Makes the inode64 mount option a little
useless on linux to be honest.
>
> > The theoretical maximum for XFS without the limits imposed by the
> > Linux block and vfs layers is probably on the order of 2^6x which
> > is a very very big number and would probably require the output of
> > several major drive vendors for a long time to construct, around 115
> > million 160G drives by my calculation!
>
> Great!
> How about 64bit linux? (e.g. XFS for IA-64 linux)
Well, so far that does not help much, it all depends on what the sizes
of the basic types generated by gcc are on those platforms, I do not
have a list for all platforms, but so far as I know, ia64 for example
is still stuck with 32 bit inodes and 2Tbyte block devices. However,
a patch was just floated on the ia64 list to do 64 bit device access
and other components of that work have appeared in the past.
Steve
--
Steve Lord voice: +1-651-683-3511
Principal Engineer, Filesystem Software email: lord@xxxxxxx
|