Re: Better data balancing over multiple devices for raid1/10 (read)

[Timofey Titovets <nefelim4ag@gmail.com>]

вт, 18 дек. 2018 г. в 19:43, Zdenek Kaspar <zkaspar82@gmail.com>:
>
> Hello, regarding Wiki/Idea pool:
> Better data balancing over multiple devices for raid1/10 (read)
>
> By looking at [1] (which I currently use) and [2, 3, 4]
> is there agreement to solve this?
>
> [1] https://patchwork.kernel.org/patch/10681671/
> [2] https://patchwork.kernel.org/patch/10403299/
> [3] https://patchwork.kernel.org/patch/10403303/
> [4] https://patchwork.kernel.org/patch/10403301/
>
> TIA, Z.

I think none of that,
because moreover that patches try solve different problems from my
point of view.

[1] Just try to squeeze more iops from same hardware by simple playing around
with some rules of queueing theory.

[2..4] Try add some duct tape to control over raid1 mirror read for
automatization and testing purposes
(i'm really prefer use sysfs knobs for that type of things).

If you try to read some mailing list conversations about above
patches, you will see
what currently we have no consensus about that solutions.
With no consensus with some majority of devs no progress will be made.

In theory that will be cool to do something similar to mdraid1, where
guys implement internal scheduling policy,
to help balance request over more complex and smart underline disk io
schedulers. That solution
can get consensus from devs POV.
But currently no one work on that, so I guess what in the near future
no agreement will be made.

-- 
Have a nice day,
Timofey.