On 27/06/17 19:41, Adrian Popescu wrote:
Hello,
I've been reading the code for a while. I'll experiment and send any
useful patches.
The technical docs would have been extremely useful in the process.
The CakeTechnical page is useful, but it lacks descriptions of the
algorithms.
Is cake still being maintained? What's the status of the cobalt branch?
Good questions.
The cobalt branch has an attractive feature of 'ingress' mode, which
from what I can tell, puts the 'bandwidth used' calculation before the
drop/no drop packet decision rather than after (as used in the normal
egress mode) - thus we add up the bandwidth that arrived rather than
bandwidth used post our drop.
I tried putting the relevant commits into master and did some simple
testing which proved ok. Shortly afterward I saw a report of increased
latency in egress mode and saw a similar thing myself - I probably
panicked a bit and reverted the 3 commits very shortly after.....but
continued running 'ingress' mode code myself. I haven't seen the issue
return and suspect a misconfig with rates on my part to be the cause.
so I suspect 'ingress' mode could go into master. For those who've
used/tested this independently it has proved useful in controlling the
torrent like behaviour of certain games providers and OS updates.
The 'lower latency priority' changes haven't obviously been a real
improvement. See the end of https://github.com/dtaht/sch_cake/issues/52
I agreed an overview of the algorithms present in the code would be
incredibly useful - I started trying to do so by reading the code a year
or so back...and the little I produced is now very out of date (it was
pre the original cobalt implementation)
Kevin
_______________________________________________
Cake mailing list
Cake@lists.bufferbloat.net
https://lists.bufferbloat.net/listinfo/cake