Alexander Wetzel <[email protected]> writes:
> I still don't understand why we don't want to use qdisc with the iTXQ
> drivers. I now just made sure we don't start using qdiscs with this
> patch to start with the least invasive approach. Anyone able to shed
> some light on that?
Because of aggregation, basically. To build an aggregate you need to be
able to pull several packets that has the same STA+TID; if all packets
on the interface are just sitting in a qdisc you have no interface to do
this.
Before switching to TXQs, drivers (the ath9k in particular) would solve
this by having another layer of per-sta queueing inside the driver,
which was just a dump FIFO that added a bunch of latency. The iTXQs
moved that up to mac80211, while doing proper queue management (flow
queueing + CoDel derived from the FQ-CoDel qdisc) at the same time; this
made the qdisc layer redundant for most purposes, which is why we
switched to noqueue by default.
-Toke