Re: [Make-wifi-fast] I have no idea how this got published...

[Pete Heist <pete@eventide.io>]

> On Jun 6, 2018, at 12:03 AM, Toke Høiland-Jørgensen <toke@toke.dk> wrote:
> 
> https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8359288
> 
> They change buffer sizes (but dynamically!) and manage to do better than
> Linux with CoDel and PIE as qdiscs on the WiFi interface. Linux 3.17.7,
> that is. This was published today…

Just after a quick read, I fail to see how just decreasing buffer sizes is going to be “better”. Better than just throwing fq_codel on the wireless interface, maybe.

In regards to CoDel and PIE, they say "both of these schemes never consider frame aggregation in the buffer sizing decision.” Isn’t the whole point of CoDel that one doesn’t need to be concerned with buffer size as long as it’s “large enough”?

> What's worse is that they actually cite our paper, with this comment:
> 
>> Recently, Høiland-Jørgensen et al. [20] also tackled bufferbloat in
>> APs with the goal to reduce latency and improve airtime fairness.
>> Their proposal was built on top of FQ-CoDel which is one of the
>> queuing disciplines in Linux. We believe that the mesh environment
>> imposes further challenges that are not addressed in this work.

That’s arm waving that really misrepresents the work without even testing it.

> I am mildly outraged…

Is there a better place to respond than adding a comment to the article?