[Bloat] SIGCOMM MIT paper: Starvation in e2e congestion control

[Bloat] SIGCOMM MIT paper: Starvation in e2e congestion control

[Dave Taht]

Perhaps it's obvious to the authors that FQ opens up new possibilities
for delay based convergence. Otherwise, pretty good:

"We prove that when two flows using the same CCA share a bottleneck
link, if the non-congestive delay variations exceed double the
difference between the maximum and minimum queueing delay at
equilibrium, then there are patterns of non-congestive delay where one
flow will get arbitrarily low throughput compared to the other. Our
theorem shows that CCAs have
to choose at most two out of three properties: high through put,
convergence to a small and bounded delay range, and no starvation."

Paper: http://people.csail.mit.edu/venkatar/cc-starvation.pdf

Article: https://news.mit.edu/2022/algorithm-computer-network-bandwidth-0804



-- 
FQ World Domination pending: https://blog.cerowrt.org/post/state_of_fq_codel/
Dave Täht CEO, TekLibre, LLC

Re: [Bloat] SIGCOMM MIT paper: Starvation in e2e congestion control

[Venkat Arun]

[-- Attachment #1: Type: text/plain, Size: 1476 bytes --]

Hi Dave,

Yes definitely, when fair-queuing is present, the onus is no longer on the
congestion control algorithm to ensure fairness. In fact, if buffer-sharing
is implemented correctly, FQ can even stand against adversarial congestion
control algorithms. We have good reason to believe that end-to-end
congestion control algorithms can provably work (i.e. achieve high
utilization, bounded delay and fairness) in the presence of FQ.

Cheers,
Venkat

On Fri, Aug 5, 2022 at 12:36 AM Dave Taht <dave.taht@gmail.com> wrote:

> Perhaps it's obvious to the authors that FQ opens up new possibilities
> for delay based convergence. Otherwise, pretty good:
>
> "We prove that when two flows using the same CCA share a bottleneck
> link, if the non-congestive delay variations exceed double the
> difference between the maximum and minimum queueing delay at
> equilibrium, then there are patterns of non-congestive delay where one
> flow will get arbitrarily low throughput compared to the other. Our
> theorem shows that CCAs have
> to choose at most two out of three properties: high through put,
> convergence to a small and bounded delay range, and no starvation."
>
> Paper: http://people.csail.mit.edu/venkatar/cc-starvation.pdf
>
> Article:
> https://news.mit.edu/2022/algorithm-computer-network-bandwidth-0804
>
>
>
> --
> FQ World Domination pending:
> https://blog.cerowrt.org/post/state_of_fq_codel/
> Dave Täht CEO, TekLibre, LLC
>

[-- Attachment #2: Type: text/html, Size: 2185 bytes --]