[LSF/MM/BPF TOPIC] Block device congestion

[LSF/MM/BPF TOPIC] Block device congestion

[Matthew Wilcox]

I wish to re-nominate this topic from last year:

https://lore.kernel.org/linux-mm/20191231125908.GD6788@bombadil.infradead.org/

I don't think we've made any progress on it, and likely won't until
everybody is forced into a room and the doors are locked.

Re: [LSF/MM/BPF TOPIC] Block device congestion

[Ming Lei]

On Mon, May 10, 2021 at 06:48:39PM +0100, Matthew Wilcox wrote:
> 
> I wish to re-nominate this topic from last year:
> 
> https://lore.kernel.org/linux-mm/20191231125908.GD6788@bombadil.infradead.org/
> 
> I don't think we've made any progress on it, and likely won't until
> everybody is forced into a room and the doors are locked.
> 

> >> Jens, is something supposed to be calling clear_bdi_congested() in the
> >> block layer?  blk_clear_congested() used to exist until October 29th
> >> last year.  Or is something else supposed to be waking up tasks that
> >> are sleeping on congestion?
> 
> Congestion isn't there anymore. It was always broken as a concept imho,
> since it was inherently racy. We used the old batching mechanism in the
> legacy stack to signal it, and it only worked for some devices.

The old batching and congestion was helpful for some slow devices, since
batching can avoid to submit IOs from different tasks concurrently to
same queue when queue is busy, so IO order in same task is maintained.

I believe Josh is struggling with this issue, and we had one offline
talk about this issue.

And there was another such old discussion too:

https://lore.kernel.org/linux-scsi/20191203022337.GE25002@ming.t460p/



Thanks,
Ming