[Babel-users] breaking the ath9k in adhoc mode with the new fq_codel implementation

[Dave Taht]

On Mon, Mar 28, 2016 at 1:03 PM, Juliusz Chroboczek
<jch at pps.univ-paris-diderot.fr> wrote:
>> scratch that. IFF_RUNNING and the ipv6 link local address does get set
>> for adhoc mode, so long as you are not also fighting with
>> networkmanager or something else flaky happens.

>> so... yea! the panda card will work in adhoc on the rpi2. It's nice to
>> know that at least one 16 dollar stick sort of works.
>
> Excellent.  (Yes, NetworkManager is evil.)

Well better integration for adhoc + network manager would be nice.

/me hides

I may have had networkmanager interfering with my life in a couple
other tests along the way, after documenting everything that you can
do wrong,
I'll go write up a matrix of how to get it right( er) later.

>> the throughput is abysmal, however. (7-9mbit in my test)
>
> What about latency?  Both without load and under load?

without: 3-36ms

with load: 500ms-3sec for one way traffic.

http://www.taht.net/~d/panda/3secofdatabackedupat10mbit.png
 (route flap? interference? rate control?)

more flent data in that dir.

We are nowhere near being able to handle usb wifi as good as the
ath10k fq_codel proof of concept was. (we never figured out how to
make usb ethernet better either) My goal was to get a bunch of itty
bitty boxes to generate loads with while we improved the ath10k,
ath9k, and got distract

(  for those that haven't seen the glory of fq_codel at the right
layer of the wifi stack, see:
http://blog.cerowrt.org/post/fq_codel_on_ath10k/ )

>
> -- Juliusz
>
>
>