filesystem limitation

[Marcus Brinkmann]

Hi,

we should revive the 1gb filesystem limit discussion.  I am not sure
if anybody ever will get to fix that ;) but it's the most obvious
and annoying limitation in our system, and people are asking often about
it.  Some volunteer to work on it (or at least say so), but I can't even
give them clear directions because last time it was discussed we had two
opposing views on how to do that.

Another issue is ext3 support, the reason I am raising it is that I
don't know if it is related or not.  Is implementing ext3 support
independent from fixing the 1gb limit or does it involve heavy pager
issues?

Back to the pager stuff.  We need to fix all places in the filesystem
where data from the disk is accessed with byte offsets.  For these, we
need to use blocks, so we don't spill over the number type.
We also need to change the libdiskfs disk pager stuff to deal with
the new model (whatever it is) and change the catch_exception stuff
accordingly.  What I am not sure is if these changes to libdiskfs can be
done independently from the actual solution used to access the disk
metadata, or not.  I am not even sure what the changes to libdiskfs
would look like.  (maybe I am reading too much into that and these parts
of lidiskfs would not be used anymore in a large filesystem).

The two opposing models I think I have summarized earlier here.  The
question is, how do we come to a decision?  I think the current status
is that Roland thinks the hierarchical locking model is feasible, while
Thomas says he has attempted it in the past and it wasn't.  Maybe Thomas
can give us a clearer idea on what the disadvantages and probles were?

Thanks,
Marcus