Re: [PATCH] blk-throttle: support io merge over iops_limit

[Yu Kuai]

Hi,

在 2025/03/10 23:53, Tejun Heo 写道:
> Hello,
>
> On Mon, Mar 10, 2025 at 10:16:46AM +0800, Ming Lei wrote:
> ...
> > > Yes, but it's a litter hard to explain to users the differece between IO
> > > split and IO merge, they just think IO split is the numer of IOs issued
> > > to disk, and IO merge is the number of IOs issued from user.
> >
> > Here it is really one trouble.
> >
> > a) Sometimes people think IOs wrt. IOPS means that the read/write IO
> > submitted from application, one typical example is `fio`.
> >
> > b) Sometimes people think it is the data observed from `iostat`.
> >
> > In a), io merge/split isn't taken into account, but b) does cover io
> > merge and split.
> >
> > So question is that what is the correct way for user to observe IOPS
> > wrt. iops throttle?
>
> If we could go back in time, I think the right metric to use is
> hardware-seen IOPS as that's better correlated with the actual capacity
> available (the correlation is very flawed but still better) and the number
> of issued bio's can easily change depending on kernel implementation
> details.

Yes, I agree the right metric is to use b), which cover IO merge and
split(and also filesystem meta and log).

> That said, I'm not sure about changing the behavior now. blk-throtl has
> mostly used the number of bios as long as it has existed and given that
> there can be a signficant difference between the two metrics, I'm not sure
> the change is justified at this point.

If we really concern about the behavior change, can we consider a new
flag that can switch to the old behavior? We'll see if any user will
complain.

Thanks,
Kuai

> Thanks.