Enable the "64bit" feature flag for filesystems over 16TB (2^32
blocks, really). Otherwise, the user needs to specify "-t ext4"
explicitly for such filesystems, or otherwise specify all of the
mke2fs features explicitly to format huge filesystems.
Porting the LU-255 patch to b1_8, I also wanted to make sure that
the new default inode ratio for OSTs was not too aggressive. While
Lustre filesystems generally have average file sizes over 1MB, in
some limited number of cases the user filesystem statistics sent to
lustre-discuss had filesystems with an ratio of 600kB/inode region.
While the average FILE size may be > 1MB/file, if the default LOV
striping is 2 or 4 stripes/file then there may be a large number of
zero-length inodes on the OST that will be allocated and need space.
Reduce the default inode ratio to previously changed in commit
eb012d4a10208b26c2d3e795a90f1bb07dde6d91 from 1MB/inode down to
512kB/inode for filesystems between 4TB and 16TB, and only go up to
1MB/inode for filesystems over 16TB. Since there are no existing
filesystems that large, there is no firm expectation for how many
inodes should be created for such filesystems, so this is safer.
Change-Id: I92139755468c5f58830849ebd44171951f8460e4
Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-on: http://review.whamcloud.com/996
Tested-by: Hudson
Reviewed-by: Yu Jian <yujian@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>