[ath9k-devel] Is there a way to connect pairs of wifi cards to achieve full duplex

[Felix Fietkau <nbd@openwrt.org>]

On 2013-06-20 8:13 PM, David Goodenough wrote:
> On Thursday 20 Jun 2013, Felix Fietkau wrote:
>> On 2013-06-20 12:26 PM, David Goodenough wrote:
>> > On Wednesday 19 Jun 2013, Ben Greear wrote:
>> >> On 06/19/2013 03:56 PM, Adrian Chadd wrote:
>> >> > .. just keep in mind that adjacent high power transmitters can
>> >> > actually leak enough RF to trigger ADC saturation and thus the device
>> >> > may actually not try to decode anything.
>> >> > 
>> >> > Thus, whilst your TX is TXing, the RX side may be unhappy. :-)
>> >> 
>> >> We've had decent multi-NIC throughput when there is a mostly-solid
>> >> aluminium chassis plate between the NICs, and when one is on 2.4
>> >> and the other is on 5Ghz.
>> >> 
>> >> Pretty much anything else is pushing your luck though :)
>> >> 
>> >> Ben
>> > 
>> > The only place I have noticed that do this with wifi kit is Microtik
>> > who say they can do setups like this - but as usual with them there is
>> > no indication of how it is done under the covers.
>> 
>> They're doing it with hacked up proprietary protocol modifications.
>> 
>> So I get why you would want to do combine two links to get full-duplex.
>> But why would you want to mess around with things like ACKs?
>> 
>> - Felix
> Most of my links are quite long, so removing any turnaround had got to
> be a good thing (hasn't it?).  I presume that every time the link turns
> around you have to turn off the receive path (or at least disconnect it)
> and then power up the transmitter, and allow time for the receiver at 
> the other end to sync.  Then at the receiving end if it was transmitting
> before (this is point to point) it has to shut down the transmit path
> in an orderly fashion so that it remains loaded while active. Surely that has 
> to take a significant time?
The rx/tx switching time is very short, short enough to be
insignificant. The inter-frame duration, slot time, etc. are
significantly longer, but aggregation effectively cuts it down to a
fraction relative to the number of packets.
With a lot of changes to the software it would be possible to disable
the on-chip ACK functionality, retransmission, etc.
However, I think the disadvantages outweigh the advantages, mainly due
to the extra latency of processing lots of things in software that are
currently accelerated by the hardware.
The increase in complexity is not particularly nice either.

- Felix