Re: [PATCH] block: Improve shared tag set performance

[Keith Busch <kbusch@kernel.org>]

On Fri, Oct 20, 2023 at 09:17:11AM -0700, Bart Van Assche wrote:
> On 10/19/23 21:41, Christoph Hellwig wrote:
> > On Wed, Oct 18, 2023 at 11:00:56AM -0700, Bart Van Assche wrote:
> > > Note: it has been attempted to rework this algorithm. See also "[PATCH
> > > RFC 0/7] blk-mq: improve tag fair sharing"
> > > (https://lore.kernel.org/linux-block/20230618160738.54385-1-yukuai1@huaweicloud.com/).
> > > Given the complexity of that patch series, I do not expect that patch
> > > series to be merged.
> > 
> > Work is hard, so let's skip it?  That's not really the most convincing
> > argument.  Hey, I'm the biggest advocate for code improvement by code
> > removal, but you better have a really good argument why it doesn't hurt
> > anyone.
> 
> Hi Christoph,
> 
> No, it's not because it's hard to improve the tag fairness algorithm
> that I'm proposing to skip this work. It's because I'm convinced that
> an improved fairness algorithm will have a negative performance impact
> that is larger than that of the current algorithm.
> 
> Do you agree that the legacy block layer never had any such fairness
> algorithm and also that nobody ever complained about fairness issues
> for the legacy block layer? I think that's a strong argument in favor of
> removing the fairness algorithm.

The legacy block request layer didn't have a tag resource shared among
multiple request queues. Each queue had their own mempool for allocating
requests. The mempool, I think, would always guarantee everyone could
get at least one request.