[Ecn-sane] for the record...

[Ecn-sane] for the record...

[Dave Taht]

I don't understand the "immediately forgotten" part of this plot. For
pure non-ect traffic should have hit 50% drop and more or less stayed
there. With a mix of ect and non ect traffic we should have hit the
drop_batch facility a lot and never got to a right drop rate.

https://l4steam.github.io/overload-results/

-- 
I tried to build a better future, a few times:
https://wayforward.archive.org/?site=https%3A%2F%2Fwww.icei.org

Dave Täht CEO, TekLibre, LLC

Re: [Ecn-sane] for the record...

[Bob Briscoe]

Dave,

I'm replying on the SANE list where this question was posted, but this 
is not related to ECN - it's about bugs in the way CoDel recovers its 
dropping mode state. Pls move it to a more appropriate list if you want. 
See [BB]

On 25/02/2022 21:22, Dave Taht wrote:
> I don't understand the "immediately forgotten" part of this plot. For
> pure non-ect traffic should have hit 50% drop and more or less stayed
> there. With a mix of ect and non ect traffic we should have hit the
> drop_batch facility a lot and never got to a right drop rate.
>
> https://l4steam.github.io/overload-results/
>

[BB] Last Oct I wrote these slides, initially as a notepad for myself, 
to work out why 'count' was being forgotten when CoDel re-entered 
dropping mode.
     https://bobbriscoe.net/projects/latency/CoDel-delta-bug.pdf
I fixed the slides up a bit to try to make them understandable by others.
And I've just added the last slide to put this bug in the context of the 
larger design flaw in the control law itself.

I'm not saying my remedies to the bugs are necessarily correct, but at 
least the slides explain what the bugs are.
I can't understand how no-one has noticed this bug before tho'.


Bob

-- 
________________________________________________________________
Bob Briscoe                               http://bobbriscoe.net/