Hi Andreas, Ted,
I've been looking at the uninitialized block group patches to see how
they overlap with the uninitialized inode tables that I was looking
into for FLEX_BG and I think I can achieve the same thing using those
patches. One of the things I wanted to achieve with uninitialized
inode tables was to pack inode for the same META/FLEX_BG and allocate
new inode tables as we ran out a pre-initialized inodes. This is
something that applies to FLEX_BG since the layout of the meta-data has
changed.
It seams that the only thing that would prevent the uninitialized block
group patches from doing this is ialloc.c unawareness of
META_BG/FLEX_BG. It seems like this is a good excuse to make inode and
block allocation algorithms aware of the new meta-data layout.
I could add META_BG/FLEX_BG awareness and see how the
allocation/initialization of inode changes and also see what sort of
performance impacts we see by these changes.
Thoughts?
-JRS