[Bloat] exploring a strangeness in a modern tcp and kernel

[Bloat] exploring a strangeness in a modern tcp and kernel

[Dave Taht]

I have been off trying to get good benchmarks of the new mikrotik 7.1
release, which has a working version of cake and fq-codel. Got a very
enthusiastic tester running the "xanmod" kernel which among other
things, has bbrv2 on by default.

Incredibly long, often confusing, and ever growing thread over here:

https://forum.mikrotik.com/viewtopic.php?p=897719#p897629

News so far:

cake doesn't crash - that was my basic objective!! - as it would crash
in mikrotik 7.0. It works great on our basic suite of tests up and
down...
Tons of flent plots vs a vs fq-codel, a fifo, cake, and sfq (sfq is
the goto default in the mikrotik world)
First ever test of BBRv2 vs cake that I know of (we moved back to
cubic later). See pics.
First time I've ever looked at how a bonded dsl connection works

However, after finally getting a packet capture, the *stumper* is that
tests using 8 or more flows actually get capped throughput well below
what the link is capable of, and I have no idea of the cause, as yet.
Up until now I was thinking the aqm was to blame, but...




-- 
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