Hi, Xiaohui,

On 28.02.2020 10:07, xiaohui li wrote:

hi Kirill Tkhai:

I agree with your idea very much.
I had also implemented a similar fallocate interface with the unique
flag which call tell ext4 filesystems to allocate the special free
physical extents.
I had done the same work as your fallocate2 work last year.

but i think it has modified the core ext4 physical blocks allocator a
lot, so the ext4 community may not accept it.
so I didn't share it openly.

but i think this fallocate2 interface just same as my work is also
very useful in android mobile phone world.

today android phone has large capacity memory and storage, just the
same as a computer.
and current days, customer treat phone and make full use of it by just
the same way as they treat computer in past days.
so after install many software and unstall them for many times in
android phone,  the ext4 physical layout on disk has become more
fragmented
(the number of extents per group is large).
at this moment, when run a sequential write workload, you will find
the sequential write performance is very bad.

but after do defragment work on some fragmented ext4 groups, and then
run the same workload again, the sequential write performance has
improved greatly.
and the fallocate2 interface is a necessary component of this
defragment work shown above.

so i also think this fallocate2 interface is an important and useful
tools for ext4 filesystems.

I hope fs people in their comments will help to formulate an interface,
which is suitable fs philosophy and for our necessities.

Do you have your defrag util hosted on some of github repositories?

Kirill