Ext4 Developer Interlock Call: 10-01-07 Minutes
Attendees: Mingming Cao, Aneesh Veetil, Dave Kleikamp, Akira Fujita, Ted Ts'o, Jose Santos, Avantika Mathur, Jean-Pierre Dion
Minutes can be accessed at: http://ext4.wiki.kernel.org/index.php/Ext4_Developer%27s_Conference_Call
Multiple Block Allocation:
- Main outstanding issue is still need for better documentation of the code.
- Aneesh has marked the places that need explanation with fixme, and Alex will be adding the needed comments. If this gets done quickly, we can try to push this feature to 2.6.24
- Aneesh has asked for reccommendations for a benchmark to run which will test the mballoc allocator. Mingming suggested using dd and dbench, as was done in the past for basic mballoc patches. To test the in-core preallocation, kernel untar tests. Look at the file fragmentation after running the test.
Large Block Patches:
- Mingming will post these patches to lkml for review
Uninitialized block groups:
- the uninitialized block group patch were posted to lkml and are in the -mm tree.
JBD-stats through procfs patches:
- Ted will look at these patches and see if they are ready to push upstream
- This function is only used in multiple block allocation patches. It would not be a good idea to post this before mballoc, since the function would be unused and may be cleaned up later.
- There has been an lkml thread about these patches.
- We have and approach that works for ext4, implemented at the vfs level, but unless we can prove it can work for other filesystems, it will not be accepted.
- Christoph Hellwig has commented that these patches will not work for XFS.
- Mingming will try the patches with ext2/3, and Shaggy will try to find time to test with JFS.
- If this approach works for other filesystems, we can try to push to mainline as a placeholder.
Truncate Mutex Patch:
- There is a performance concern with these patches. Mingming suggested doing some testing to see what the actual performance regression is.
- Ted suggests breaking into two patches:
- 1st Patch - Change the name of the mutex to block_mapped_mutex and update comments to reflect the usage of the mutex
- 2nd Patch - Change to a read/write mutex
- Those patches we would still like to see in 2.6.24 need to be pushed very quickly
- We should avoid having ext2/3 patches in the ext4-patch-queue and git tree.
- Some of the patches that we just send to lkml for review end up being pulled into the -mm tree, then Ted pulls them to the ext4-git tree, and Andrew must remove from -mm. When we post for comments, we should use [PATCH,RFC] to avoid this.
- Mingming suggested having a patch queue for the ext4-git tree for easier review. Ted is also considering having a pu branch for the git tree. Ted will think about what will work best.
- In the ext4-patch qDescription of patch include which version of -mm the patch is in