Ext4 Developer Interlock Call
October 22, 2007: Meeting Minutes
Attendees: Mingming Cao, Andreas Dilger, Eric Sandeen, Dave Kleikamp,
Jose Santos, Aneesh Veetil, Valerie Clement, Avantika Mathur
- There has been discussion on linux-ext4 about overflow in ext2 with 64
KB block size. Discussed adding an incompat flag to this feature and
concluded that it is too late to do this; it will be treated as a bug fix.
Patch Status: patches that were not picked up to mainline in the recent
pull:
- journal_checksum patches - the fsstress kernel oops is not yet
resolved, Avantika will be working on this.
- i_version: Jean-Noel recently sent out an updated version which adds a
lock every time i_version is updated. Still waiting on more review
- mballoc - target for next kernel release. Remaining work:
- Aneesh plans to run some performance test and include results in
the patch description.
- Waiting for Alex to address the fixme's that aneesh has added to
the code, for more clarification.
- block groupdescriptor sync with e2fsprogs:
- Aneeesh recently posted patches which have been added to the patch queue.
E2fsprogs
- Ted has added new branches the the e2fsprogs git tree
- master: current stable version
- next: patches expected to go into the next stable version
- pu: proposed update; patches that are in preliminary review and
test phase
- E2fsprogs patches should probably be submitted against the 'next'
branch of the git tree.
- Aneesh was looking for patches to support mke2fs of large
filesystems. Valerie's old patches don't apply cleanly.
- Documentation in the kernel tree; ext4.txt is out of date and needs to
be updated.
Performance:
- Andreas had asked Chris Mason to add ext4 to his BTRfs benchmarks.
Chris Mason reported that ext4 is doing poorly on a test which performs
kernel untar and compile on 20 kernels in a row. The performance of
ext4 on reads is poor compared to ext3.
- Eric will talk to Chris and get all of the details of the tests.
- Eric suggested enabling delalloc,mballoc mount options by default.
The dellaloc option can then be disabled if the fs is not in writeback
journaling mode.
Hello,
Sorry for not taking part in the call but I was just on my way from NY
to Prague...
> Ext4 Developer Interlock Call
> October 22, 2007: Meeting Minutes
>
> Attendees: Mingming Cao, Andreas Dilger, Eric Sandeen, Dave Kleikamp,
> Jose Santos, Aneesh Veetil, Valerie Clement, Avantika Mathur
>
> - There has been discussion on linux-ext4 about overflow in ext2 with 64
> KB block size. Discussed adding an incompat flag to this feature and
> concluded that it is too late to do this; it will be treated as a bug fix.
Hmm, why adding an incompat flag? Obviously there's no harm in doing
that but the large block size itself should be enough to protect old
kernels/utils touching it, shouldn't it?
> E2fsprogs
> - Ted has added new branches the the e2fsprogs git tree
> - master: current stable version
> - next: patches expected to go into the next stable version
> - pu: proposed update; patches that are in preliminary review and
> test phase
> - E2fsprogs patches should probably be submitted against the 'next'
> branch of the git tree.
BTW: We still don't have appropriate patches for large block size in
e2fsprogs. I guess I should ping Ted about merging the patch once more.
Honza
--
Jan Kara <[email protected]>
SuSE CR Labs