I am encountering an issue, that wireless devices appear to only get one Tx stream on a RBcAPGi-5acD2nD cAP ac running Firmware 6.48.5.
I recently performed a few throughput tests using iperf3 (client<->cAP ac<->iperf3-server) and was surprised, that I could only achieve
Rx over 2 minutes: 310-320 Mbits/sec
Tx over 2 minutes: 40-50 Mbits/sec
After various tests (including the cable connecting between switch and router (which is fine with 920+ Mbits/sec)) I found, that the cAP ac uses a
Rx Rate of 866.6Mbps-80MHz/2S/SGI
(device approx. 3m away and in sight), but only one stream in the other direction:
Tx Rate 433.3Mbps-80MHz/1S/SGI.
This behavior occurs both for
Intel Dual Band-Wireless-AC 8265 (2x2) (Q1/16)
Intel Wi-Fi 6 AX2/01 (2x2) (Q2/19)
wireless adapters.
However, when connected to a different 802.11ac access point (Qualcomm Atheros QCA9880) a reasonable net throughput (Rx over 2 minutes: 380-390 Mbits/sec, Tx over 2 minutes: 450-460 Mbits/sec) in both (!) directions can be achieved.
Also: this behavior can be observed in stand-alone (testing for me) and CAPsMAN-/controller-based (my default) mode.
Any ideas what the reason is?
Is this a configuration or a hardware (access point) issue?
If you need further information, please let me know.
I don’t think noise is an issue, as there is no other network on the same (5GHz-)channel (not even an overlap) and as I disabled the wifi connection on all devices (except the one used for the test).
Throughput tests show corresponding results: device to cAP ac above 400 MBit/s, cAP ac to device always under 200 MBit/s, most of the time even lower than 100 MBit/s
Summary: same result, great performance (> 400 MBit/s) when sending data from the device to the cAP ac, poor performance when downloading data (cAP ac to device).
Another observation: when I move futher away and other devices are connected, too, the Tx Rate changes to for example 390Mbps-80MHz/2S/SGI. However, for Tx (cAP ac to device) 866.6Mbps-80MHz/2S or anything near that is never achieved.
Any ideas how to achieve acceptable data rates/throughput for Tx?
using Intel Centrino Ultimate-N 6300 (802.11 a/b/g/n)
Note: these tests were performed in a non-optimal environment with approx. 3m-4m distance to the access point and other devices connected.
The test with the AX201 shows the usual behavior:
Rx: 320-390 MBit/s with two streams
Tx: 90-190 MBit/s with only one stream
(really? 90 MBit/s?)
The results with the “old” Intel 6300 n-only card were very interesting:
Both Rx and Tx used two streams continuously and “max-ed out” what n-wifi offers, RouterOS continuously showed 300 MBit/s or values near that in the Registration Table for the device. The actual usable throughput correlated to that.
As I have not mentioned it: the tests were performed using iperf3 with 10 threads over three minutes. iperf3 runs on my router which is directly connected to the access point. The iperf3-client runs on the device (various laptops) that I use for the test.
Could there possibly be an 802.11ac issue with the cAP ac?
Am I really the only one observing this behavior?
Thank you very much for the suggestions and links.
Unfortunately, this did not help …
Thank you for posting your configuration.
My configuration looks similar.
The only differences (besides adapter specific availability of some properties):
Mixed Mode Protection - My setting is “RTS/CTS Enabled”, I didn’t change it, since I am not using 802.11b oder 802.11g in the 2.4 GHz band.
Preferred Band - I kept “No Preference” as the network uses 5 GHz only
Roaming Aggressiveness - Kept “Medium” as roaming is not applicable with only one access point
Throughput booster - According to Intel’s documentation this only effects the upload from the device to the access point. That was never an issue.
However, I did change the “MIMO Power Save Mode” from “Auto SMPS” to “No SMPS”.
Good news: when the connection is idle (!) Tx Rate 520Mbps-80MHz/2S/SGI and Rx Rate 866.6Mbps-80MHz/2S/SGI are displayed. Appearently the “MIMO Power Save Mode” affects the usage of the antennas.
Bad news: When I start the throughput test the displayed data rates massively drop, e.g. to values
Rx 40.5Mbps-40MHz/1S
Tx 234Mbps-80MHz/2S
resulting in a net throughput as low as
Rx 31 MBit/s
Tx 103 MBit/s
Therefore I changed the setting back to “Auto SMPS”.
In addition I did tried the same again using a TP Link Archer T3U Plus USB WLAN Stick as I have not re-tested this device with RouterOS 7.2.1.
Rx is ok as usual: most of the time 780Mbps-80MHz/2S/SGI (sometimes even 866.6Mbps-80MHz/2S/SGI), resulting in 304 MBit/s.
Tx still uses one stream only: most of the time 325Mbps-80MHz/1S/SGI resulting in 183 MBit/s.
Note: Tests were performed after a restart of the device (laptop) and the cAP ac, no other devices connected to the access point, approx. 2m distance to the access point with direct line of sight. No other SSIDs on the same channel. Note: Throughput tests were performed using iperf3 over 60 seconds with 10 threads parallel. iperf3 runs on my router which is directly connected to the access point. The iperf3-client runs on the device (laptop).
I still believe (but do not know), that there is some issue (eventually only a configuration) with the cAP ac, as
using a different access point (Qualcomm Atheros QCA9880, see my post Wed Oct 27, 2021 11:35 pm) the problems mentioned do not occur
the issue is the same with three different network adapters (Intel 8265, Intel AX201, TP Link T3U Plus)
After reading all the interesting posts here … just thinking …
"SMPS is a problem if enabled (reduces the max # receive antenna to just 1 on the client)
“line of sight” and close proximity are not the good conditions for spatial streams (MIMO) operation. Multipath is a must. See https://www.ieee.li/pdf/viewgraphs/introduction_to_wireless_mimo.pdf
(Having the client In the backplane direction of the cAP, the spreading of the antenna might be sub-optimal, (seen in a small angle))
if RTS/CTS is needed for “auto SMPS” then it should be enabled on the AP.
Another observation: when I move futher away and other devices are connected, too, the Tx Rate changes to for example 390Mbps-80MHz/2S/SGI. However, for Tx (cAP ac to device) 866.6Mbps-80MHz/2S or anything near that is never achieved.
reduced speed is a normal reaction of the wifi algoritms if packets don’t make it at the higher speeds. (7 are the HW retries without consequences, then the interface rate is stepped down for every failure until it works. So hey, with “390Mbps-80MHz/2S/SGI”, 2S does work, but the MCS encoding had to be strongly reduced to MCS04 or 16-QAM for the noise or interference.
52-64 are 5 GHz non-overlapping channels.
What is meant here? 5260Ceee, 5280eCee, 5300eeCe , and 5320eeeC use one and the same 80 MHz channel. If 5300Ceee would have been selected, then there would be serieus interference with the previous channels.
This is my understanding, too.
However, I changed the settings back as - possibly due to a different issue - I could not observe an improvement.
Today I switched to “no SMPS” again, see below.
Interesting presentation!
Long story short: I would still expect something at least close to 866Mbps (using two streams) if there are no disturbances around.
Where did you see/read, that RTS/CTs may be needed for “Auto SMPS”?
There is really noting around … :-S
I have not explicitly set the Extension Channel.
You are right. For some reason I mixed something up with the non-overlapping 80 MHz channels.
Don’t know, what I was thinking.
For MIMO there must be some diversity between individual streams in order for receiver to distinguish between individual streams. In our physical universe, there are two ways of achieving needed diversity:
the most straight forward way is polarization diversity. There are only two orthogonal polarization planes so it’s only good for 2x2. It’s achieved by having a pair of antennae set at right angle between them, professional antennae actually have the two one over another like a cross. Multipath makes things worse in this case, polarity generally changes at every reflection. So for clear LOS cases, MIMO 2x2 with polarization diversity should work close to ideally. Ideal orientation of receiving antennae helps, so one xan try to rotate antennae around the direction pointing at the transmitter to match Rx polarization planes with Tx polarization planes as good as possible.
__
2. spatial diversity, which means that antennae are set certain distance apart. It works fairly for higher orders of MIMO, but to do it properly, distance between any pair of Tx or Rx antennae has to be at least few wavelengths. For transmitters operating at 2.4 GHz, wavelength is around 12 cm, so antennae should be separated at least 30-50 cm apart. Which obviously doesn’t happen with any consumer electronic device (except for carefully engineered laptops). Wifi at 5GHz is a bit better with this regard, wavelength is approx 2.2 to 2.5 times shorter which makes construction of better antenna much easier. BTW, mobile networks sometimes use spatial diversity even on low bands … with frequencies as low as 700MHz this means distance between antenna of at least 3 meters … but when looking up at tower keep in mind that panel antennae used for that frequency are 2.5m to 3m high.
Now, with spatial diversity multipath can help because (multiple) reflections can make actual distance difference between Tx antennae appear larger at receiver … but can make it vanish as well. So the effect can vary a lot even with small receiver movements. OTOH multipath can cause receiver to sense increased interference if it can’t cope with multipath properly (it’s not exactly trivial and I can imagine low cost receivers lacking DSP processing power and/or shitty DSP software burp in some cases) and in that case LOS SISO might give better results.
The receiver can have issues in both cases of diversity, also in case of polarization diversity if polarization planes of Tx and Rx antennae are rotated at 45°. And note that transmitter doesn’t have any issues in such cases … so large difference in performance in both directions might indicate lower quality of receiver (the direction with lower throughput) if the rest of processing doesn’t make bottlenecks (e.g. CPU on either transmitting or receiving device, my experience is that transmitting is harder on CPU than receiving).
