[LEDE-DEV] Bug in ath9k coverage distance
Koen Vandeputte
koen.vandeputte at ncentric.com
Tue Jun 7 04:02:45 PDT 2016
Hi all,
Chaoscalmer has a lot less wifi bandwidth when coverage class is increased.
Even when 2 devices are only a few meters apart.
In Barrierbreaker release, the rates were a lot higher in this usecase.
When doing Iperf benchmarks (~ 30 sec);
* The barrierbreaker 14.07 release is able to hit ~53Mbit from start
to finish
* Chaoscalmer starts at ~6Mbit/s, than slowly goes up to ~20Mbit/s.
But starts at 6 again when the test is repeated
Hardware used for testing:
2x Gateworks Laguna CNS3xxx + AR9280 (Ubnt sr71-15)
2x Gateworks Ventana IMX-6 + AR9300 (WLE350NX)
Branch tested:
5770678122e76079debd6eeeb118f32e3fc78b84 (Master @ June 1st 2016)
Following setup:
* 2 identical devices
* IBSS mode using wpa_supplicant WPA2
* 2 meter apart
* Free 5GHz channel (confirmed using spectrum analyzer)
* HT20 (HT40 has same behaviour)
* Coverage class 32 (up to 14400m)
Iperf using BarrierBreaker build: ~52Mbit/s
Iperf using ChaosCalmer build: ~6Mbit/s
* Coverage class 16 (up to 7200m)
Iperf using BarrierBreaker build: ~53Mbit/s
Iperf using ChaosCalmer build: ~12Mbit/s
* Dynack
Iperf using BarrierBreaker build: ~55Mbit/s
Iperf using ChaosCalmer build: ~53Mbit/s
Steps taken without success:
* Upgraded wpa_supplicant to latest trunk
* Checked ath9k code hw.c & recv.c (Acktime, slottime, etc. But code
doesn't differ between BB release and Master branch)
* Disabled any possible background scanning
* Patches deleted - In delta to BB release: (in kernel/mac80211/patches)
--> 326-ath9k-make-NF-load-complete-quickly-and-reliably.patch
--> 330-mac80211-minstrel-Change-expected-throughput-unit-ba.patch
--> 334-mac80211-fix-wiphy-supported_band-access.patch
--> 335-mac80211-minstrel_ht-set-A-MSDU-tx-limits-based-on-s.patch
--> 336-mac80211-minstrel_ht-set-default-tx-aggregation-time.patch
--> 337-mac80211-minstrel_ht-fix-a-logic-error-in-RTS-CTS-ha.patch
--> 343-mac80211-minstrel_ht-improve-sample-rate-skip-logic.patch
Not yet checked:
* Minstrel_ht
If more information is required, please let me know.
Thanks,
Koen Vandeputte
nCentric Europe
More information about the Lede-dev
mailing list