Ext4 Developer Interlock Call: 11-29-06 Minutes
Attendees: Mingming Cao, Eric Sandeen, Dave Kleikamp, Jean-Pierre Dion,
Val?rie Cl?ment, Jean-Noel Cordenner, Takashi Sato, Ted Ts'o, Avantika
Mathur
- Preallocation:
-- Suparna is has started working on implementation and will
temporarily use ioctls as interface.
-- Discussed XFS implementation details posted by Eric, why there are
two methods, one zeroing out blocks, and the other setting flag in
preallocated extents.
-- Eric will work on a proposal for preallocation implementation using
posix_fallocate and fcntl
- Nanosecond Timestamp:
-- Reviewed status of Patch from Andreas Dilger in June
-- Before moving forwars, we need to address design question of
whether nanosecond timestamps should be supported in older filesystems
with 128 byte inodes. Will this feature be backported to ext3?
-- Ted will reopen this issue on the mailing list
-- Current plan is to only implement finer granularity timestamps for
larger inodes, unless there is someone who feels it is critical to have
them in current inodes.
- Defragmentation Patches:
-- Takashi is working on addine improvements including support for
quotas.
-- Will release new patch set in early December
- e2fsprogs: No updates to e2fsprogs since last week.
-- New WIP released week ago; includes some recent patches including
struct defs for new features
-- Ted is still working to merge extents patches and will report
progress next week
- New Wiki: ext4.wiki.kernel.org Still a lot of work to be done to set
it up.
-- We will keep all interlock agendas and minutes here
-- Will create ext4 roadmap with details on implementations for each
feature
- Big Block Group patches:
-- Val?rie posted patches last week
-- Need a clearer description of why this feature is needed and what
the benefits are
-- Decide whether it is a necessary feature and if all large
filesystems should be using it.
-- Ted will reply to Val?rie's patches asking about specific benefits
and requirements.
-- Current thoughts on benefits of large block group size:
o increase file system size
o reduce file fragmentation because less extents are needed for
large files
o Reduce number of block descriptors, which reduces mkfs and fsck
times for large filesystems
>>>>> Avantika Mathur LTC (AML) writes:
AML> o reduce file fragmentation because less extents are needed for
AML> large files
say, current limit for a single extent is 2^15 blocks or 128MB.
after regular mke2fs, a group has ~125MB free.
thanks, Alex