Hi Maximilian,
I read the following:
"D. Other variants of fair queuing
We also evaluated the performance of our fair queuing detection on a bottleneck
managed by fq codel [5]. We chose a default target queuing delay of 10ms
following Apple’s implementation3 because we argue that Apple probably spent a
considerable amount of time fine-tuning their implementation and came to the
conclusion that 10 ms work best as the default target delay."
And wonder whether you could:
a) repeat that experiment with fq_codel's defaults of 100ms interval and 5ms
target using the Linux implementation. I am not saying Apple might not have a
decent rationale for their choice, but as far as I can tell they have not
communicated that rationale. The Linux defaults however are explained
relatively well in e.g. fq_codel's IETF RFC.
b) produce some CDF plots that show the detection accuracy for the different
RTTs and rates (you can probably combine either all RTTs or al rates into one
plot)
c) maybe use signal detection theory terms to show the performance in terms
that include false positive and false negative classifications?
Regards
Sebastian
> On Jun 27, 2022, at 18:23, Maximilian Bachl via Bloat
> <[email protected]> wrote:
>
> This paper (pre-print)
> https://arxiv.org/abs/2206.10561 proposes a mechanism to monitor the presence
> of FQ continuously during a flow’s lifetime. This can be used to change the
> congestion control depending on the presence of FQ.
>
> Furthermore, the paper argues that the presence of FQ can be considered a
> congestion signal: Only if there’s congestion, FQ can be detected.
> _______________________________________________
> Bloat mailing list
> [email protected]
> https://lists.bufferbloat.net/listinfo/bloat
_______________________________________________
Bloat mailing list
[email protected]
https://lists.bufferbloat.net/listinfo/bloat
