Re: [Bloat] [Codel] ACM queue article on facebook´s "Adaptive LIFO" and codel

[Jonathan Morton <chromatix99@gmail.com>]

> On 11 Apr, 2023, at 5:12 am, Dave Taht <dave.taht@gmail.com> wrote:
> 
> I have no idea what an "adaptive LIFO" is, but the acm queue paper
> here just takes the defaults from codel...
> 
> https://twitter.com/teivah/status/1645362443986640896

They're applying it to a server request queue, not a network packet queue.  I can see the logic of it in that context, but I would also note that LIFO breaks one of Codel's core assumptions, which is that the maximum delay of the queue it's controlling can be inferred from the delay experienced by the most recently dequeued item.

Maybe it still happens to work by accident, or maybe they've implemented some specific workaround, but that paper is a very high-level overview (of more than one technology, to boot) without much technical detail.  If I didn't already know a great deal about Codel from the coal face, I wouldn't even know to consider such a failure mode, let alone be able to infer what they could do to mitigate it.

 - Jonathan Morton