are your devices all configured with on AP and one IBSS interface? Or even more AP interfaces? What is the multicast rate you used?
One thing you need to consider is that in ath9k if you run AP and IBSS at the same time (at least with OpenWRT), the distributed IBSS beaconing is disabled and each node will beacon the "normal" ~10 times per second. That means with AP and IBSS and 40 nodes, you end up with ~800 beacons per second. That might be a problem (especially if all nodes are in the same spot).
To change this, you can try:
* increase the multicast rate to e.g. 18 mbit/s (note, this may also change the batman-adv neighbor range) and verify that the beacons are sent on higher rates * disable some APs - you don't need so many anyway if they are all on the same spot * increase the beacon interval on the APs- although some clients may not like that.
.. anyway, that is all stuff you can try if you have the feeling the airtime is "polluted", so I would suggest to verify if this is the cause first before taking counter measures.
Cheers, Simon On Saturday, June 3, 2017 2:30:27 PM CEST Xuebing Wang wrote:
=> Is it a gradual thing? Ie, do you know when it starts to hang? Does it hang at 40? or gradually as you approach 40?
We do not know. What we do know is, 20 peers / neighbors work well.
Jason Toschlog in CC said that he observed similar issues with brcmfmac when peers reach 32 (16 peers work for him).
I guess I will setup 32 or 40 nodes here and try limit the max number of peers (at wpa_supplicant) for IBSS/adhoc.
On 2017年06月03日 14:16, Adrian Chadd wrote:
Well, it sounds like something you need a 40 node test network for :)
I was wondering whether it was some kind of MAC bug that was being triggered because of so many units say, overflowing the keycache or something. But if it's software encryption then maybe not. I wish I could help more, but setting up a 40 node network to test is a big challenge. :)
Is it a gradual thing? Ie, do you know when it starts to hang? Does it hang at 40? or gradually as you approach 40?
As for association limits - I thought there was a maximum station parameter somewhere...