Re: [PATCH, RFC] ext4: New inode/block allocation algorithms for flex_bg filesystems

[Eric Sandeen <sandeen@redhat.com>]

Andreas Dilger wrote:
> Ted Ts'o wrote something like the following (didnt' get original email):
>> @@ -122,6 +122,9 @@ struct ext4_inode_info {
>>  	struct list_head i_prealloc_list;
>>  	spinlock_t i_prealloc_lock;
>>
>> +	/* ialloc */
>> +	ext4_group_t	i_last_alloc_group;
> 
> Even better would be to store i_last_alloc_inode.  In the past Eric
> has demonstrated workloads that are allocating lots of inodes exhibit
> O(n^2) behaviour because the entire group bitmap is searched from the
> start each time, and that can cumulatively be very slow.  Having the
> directory start searching from the most recently allocated inode would
> make this O(n), and would not significantly alter behaviour.

A very hacky benchmark I had to demonstrate this is at

It just creates a directory tree starting at 000/ under the dir it's run
in, and times iterations of creates.

The tree is created in order, like:

000/000/000/000/000/000
000/000/000/000/000/001
000/000/000/000/000/002
...
000/000/000/000/000/fff
000/000/000/000/001/000
....

On ext3:

# ./seq_mkdirs
iter 0: 6.191491 sec
iter 1: 8.455782 sec
iter 2: 9.435375 sec
iter 3: 10.198069 sec
iter 4: 10.922969 sec
iter 5: 10.800908 sec
iter 6: 12.940676 sec
iter 7: 15.513261 sec
...

On upstream ext4:

# ./seq_mkdirs
iter 0: 5.628331 sec
iter 1: 6.581043 sec
iter 2: 6.723445 sec
iter 3: 6.567891 sec
iter 4: 5.862526 sec
iter 5: 6.462064 sec
iter 6: 7.208110 sec
iter 7: 6.549735 sec
...


I did play with saving the last-allocated position but if that's just
in-memory then it's a little odd that the first allocation will be
potentially much slower, but that's probably acceptable.  It also
wouldn't fill in gaps when inodes are deleted if you don't re-search
from the parent.  ISTR that the constant create/delete didn't cause a
problem, will need to remind myself why ...

-Eric