2007-01-10 22:08:14

by Avantika Mathur

[permalink] [raw]
Subject: Ext4 devel interlock meeting minutes (Jan. 10 2007)


Ext4 Developer Interlock Call: 1-10-07 Meeting Minutes
Attendees: Mingming Cao, Dave Kleikamp, Takashi Sato, Suparna
Bhattacharya, Jean-Pierre Dion, Jean-Noel Cordenner, Val?rie Cl?ment,
Ted Ts'o, Avantika Mathur

Minutes can be accessed at:
http://ext4.wiki.kernel.org/index.php/Ext4_Developer%27s_Conference_Call

- Conferences:
-- The Linux filesystem and storage workshop is coming up in February
-- OLS: deadline to submit abstracts is Feb 5th. Ted is thinking about a
paper on e2fsprogs. Encouraging writing in-depth technical papers on
ext4 work.

- Ted will set up a git tree on kernel.org with all of the ext4 patches;
so it is easy to access and test all of the patch sets.
- Mingming will send an email requesting everyone to resend latest
versions of ext4 patches to the mailing list, with a short description
and status including 'to dos' and outstanding issues.

Ext4 Roadmap - an update on ext4 features in progress:
These updates will also be added to the wiki page:
http://ext4.wiki.kernel.org/index.php/New_ext4_features

- Multiple Block Allocation
-- status: Patches are available and there are no format or e2fsprogs
changes. given that there are no dependencies, this feature should be
pushed to mainline soon.
-- outstanding issues: none

- Delayed Allocation
-- status: Patches are available, and have been commented on -- need
more work.
-- outstanding issues: lack of plan for 1k,2k block size and default
data=order mode, and Andrew asked for more comments.

- Preallocation
-- status: Kernel implementation and e2fsprogs changes are complete, and
have been thoroughly reviewed by Mingming and Alex Tomas.
-- outstanding issues: Need consensus on what interface should be used
(ioctl, fcntl..). What should the return value be on success, zero or
number of block allocated. Suparna will post a comment for a case where
the extent is allocated far off on the disk. Once the interface is
decided, this feature will be ready to add to the git tree.
-- Andrew Morton has mentioned adding preallocation to directories. Ted
and Mingming discussed adding readahead and reservation to directories
rather than better than preallocatin.

- Online Defragmentation:
-- status: Takashi sent updated patches last month. He will repost most
recent version to the list.
-- outstanding issues: unknown

- Break 32000 directory limit:
-- status: Andreas Dilger posted this patch. It is a simple patch with
minor e2fsrogs changes. Ask Andreas to repost patches on ext4.
-- outstanding issues: unknown

- Nanosecond Timestamps
-- status: Patches have been posted by Andreas Dilger, shouldn't need
much rework once issues are resolved.
-- outstanding issues: Should be available in 128 bit inodes, or only
in larger inodes. What effect updating the nanosecond timestamp has on
performance, excessive atime updates. Should atime be kept at ~1 second
granularity.

- Change Attribute
-- status: use i_version rather than change_attribute field for this
feature. Jean-Noel will resend out latest patch with summary and status
-- outstanding issues: unknown

- Uninitialized Groups
-- status: Feature will reduce fsck time on a large filesystem.
Currently this features is very premature and in prototype mode. Basic
functionality for testing on 64 bit is in place, but not yet useful in
production. There are no dependencies on this feature, for safety the
feature depnds on checksumming block groups. Ask Andreas to resend patches.
-- outstanding issues: prototype mode

- Journal Checksumming
-- status: feature will speed up journaling. The feature was proposed,
but there is no status on patches.
-- outstanding issues: n/a

- Undelete Feature for Ext4
-- status: Idea was proposed on the mailing list, still need more
thought on the feature.
-- outstanding issues: Should the kernel be involved with undelete

- Large file support
-- status: Increase 2Tb limit on filesize. Takashi has begun working on
this.
-- outstanding issues: unknown

- E2fsprogs work
-- Ted is working on a user issue handling > 375 million directories on
a 32 bit machine
-- Correct interfaces for extents
-- adding 64 bit API without causing ABI breakages. Might need a major
version change with the ABI breakage.

- Supporting Trillions of Files in Ext4
-- Mingming and Avantika are looking into this
-- Issues and Constraints: difficult to support more than 32 bits of
inodes, filsystem format and compatibility issues, stat interface is
only 32 bits, memory utilization of inode cache will affect performance