Page 1 of 1

hAP AC + capsman performance

Posted: Tue Apr 13, 2021 2:13 pm
by en1gm4
Looking for a little advice to try to understand how to get some better performance from a hAP AC connecting via CAPSMAN to a hEX (RB750GR2)

I thought i'd do a little advance planning pre the arrival of FTTH at home. I've been running three .11n APs (RB951G-2HnD) using capsman to date but never needed more than my 50/10 broadband allowed. Given the new install will be able to get to gigabit (or even 10) i thought some better wifi was in order.
Currently I have one home SSID and one guest SSID so capsman makes it simple to keep them apart, change configs, update all at once etc. I'd like to stay with CAPSMAN if possible.

However, and this seems like a common problem, getting good performance out of the hAP seems to be a challenge.
I have read all i can find. 5Ghz spectrum here is not at all crowded. On my phone i can only pick up one other AP and it is at -89dbm
As it is currently configured, I get an 866 PHY up and down... sitting about 2-3m from the AP. Signal is good at -45dBm

However, when i run iperf to my gigabit connected desktop I am getting a max of about 140Mbit down to the test device and often less (pixel 3a phone - 2S AC).I discovered if i reverse the test direction I have got up to 180Mbit (not consistently) . I note that many people mention the CPU on the hAP as a bottleneck but when i looked during the iperf test it was <50% loaded (And the hEX perhaps 20%).. so it doesn't seem like CPU should yet be an issue yet. I'm not using local forwarding. I tried both with and without a bridge configured on the hAP with no discernible impact.
Testing windows file transfers gives me a little lower performance.

Channel selection is auto 80Mhz and using xxxx or ceee makes no difference (I had read somewhere that xxxx could impact performance presumably if the control channel is badly located)
I'm allowing DFS channels.It takes a while to settle but works fine and seems to hold onto the channel, Other than that i have tried to change as little as possible to not complicate matters.
I'd be happy to consistently get 200Mbit or nowfbut would like to know what might be possible and how to attain it.

Everything is on 6.47.9. Country set to UK (indoor)
(side note... the hAP seemed to be unaware of its own 5Ghz antenna gain so i had to put that in manually... doesn't fill me with confidence in the code...)

Any hints or tips from those who have been through this? It does seem to be harder than it ought to be.

Re: hAP AC + capsman performance

Posted: Tue Apr 13, 2021 2:20 pm
by erlinden
Can you share: /wireless export hide-sensitive file=anynameyoulike

Re: hAP AC + capsman performance

Posted: Tue Apr 13, 2021 3:15 pm
by en1gm4
from the capsman box? (hEX)
(since /interface wireless export from there shows nothing... )
Assuming you mean from /caps-man I have included that export here
# apr/13/2021 13:11:27 by RouterOS 6.47.9
# software id = 3W03-BPBK
# model = RouterBOARD 750G r2
# serial number = 64FD026AE1C8
/caps-man channel
add band=2ghz-onlyn control-channel-width=20mhz extension-channel=disabled \
    frequency=2412 name=channel1
add band=2ghz-onlyn control-channel-width=20mhz extension-channel=disabled \
    frequency=2432 name=channel5
add band=2ghz-onlyn control-channel-width=20mhz extension-channel=disabled \
    frequency=2452 name=channel9
add band=2ghz-onlyn control-channel-width=20mhz extension-channel=disabled \
    frequency=2472 name=channel13
add band=5ghz-n/ac name=Defaullt5G
add band=2ghz-g/n frequency=2412,2431,2572,2472 name="2G basic frq"
/caps-man datapath
add bridge=bridge client-to-client-forwarding=yes local-forwarding=no name=\
add bridge=wifi_guest comment="data for guest wifi virtual SSID" name=\
/caps-man security
add authentication-types=wpa2-psk encryption=aes-ccm name=aebleskuber
add authentication-types=wpa2-psk encryption=aes-ccm name="Hillside(guest)"
/caps-man configuration
add channel=channel1 channel.control-channel-width=\
    20mhz channel.extension-channel=disabled channel.tx-power=12 country=\
    "united kingdom" datapath=datapath1 datapath.bridge=bridge distance=\
    indoors mode=ap name=skuber_CH1 security=aebleskuber ssid=aebleskuber
add channel=channel5 channel.control-channel-width=\
    20mhz channel.extension-channel=disabled channel.tx-power=12 country=\
    "united kingdom" datapath=datapath1 datapath.bridge=bridge distance=\
    indoors mode=ap name=skuber_CH5 security=aebleskuber ssid=aebleskuber
add channel=channel9 channel.control-channel-width=\
    20mhz channel.extension-channel=disabled channel.tx-power=10 country=\
    "united kingdom" datapath=datapath1 datapath.bridge=bridge distance=\
    indoors mode=ap name=skuber_CH9 security=aebleskuber ssid=aebleskuber
add channel=channel13 channel.control-channel-width=\
    20mhz channel.extension-channel=disabled channel.tx-power=10 country=\
    "united kingdom" datapath=datapath1 datapath.bridge=bridge distance=\
    indoors mode=ap name=skuber_CH13 security=aebleskuber ssid=aebleskuber
add country="united kingdom" datapath=datapath2 datapath.bridge=wifi_guest \
    mode=ap name=GuestWifi security="Hillside(guest)" ssid=Hillside
add channel=Defaullt5G channel.extension-channel=XXXX \
    channel.tx-power=17 country="united kingdom" datapath=datapath1 \
    datapath.bridge=bridge datapath.client-to-client-forwarding=yes mode=ap \
    name="skuber 5GHz generic" security=aebleskuber ssid=aebleskuber
add channel=Defaullt5G channel.extension-channel=XXXX \
    channel.skip-dfs-channels=no channel.tx-power=-30 country=\
    "united kingdom" datapath=datapath1 datapath.bridge=bridge \
    datapath.client-to-client-forwarding=yes distance=indoors \
    load-balancing-group="" mode=ap name="5ghz test " security=aebleskuber \
add channel="2G basic frq" \
    channel.control-channel-width=20mhz channel.extension-channel=disabled \
    channel.tx-power=12 country="united kingdom" datapath=datapath1 \
    datapath.bridge=bridge distance=indoors mode=ap name=\
    "Skuber 2G auto channel" security=aebleskuber ssid=aebleskuber
/caps-man manager
set enabled=yes upgrade-policy=suggest-same-version
/caps-man provisioning
add action=create-dynamic-enabled comment=Playroom master-configuration=\
    "Skuber 2G auto channel" name-format=identity radio-mac=D4:CA:6D:99:F1:6B \
add action=create-dynamic-enabled comment=Lounge master-configuration=\
    "Skuber 2G auto channel" name-format=identity radio-mac=D4:CA:6D:93:25:C9 \
add action=create-dynamic-enabled comment=Kitchen disabled=yes \
    master-configuration=skuber_CH1 name-format=identity radio-mac=\
    D4:CA:6D:BB:75:A3 slave-configurations=GuestWifi
add action=create-dynamic-enabled comment=Garden master-configuration=\
    "Skuber 2G auto channel" name-format=identity radio-mac=B8:69:F4:6B:D3:BB \
add action=create-dynamic-enabled comment="Dining Room 5Ghz" \
    master-configuration="skuber 5GHz generic" name-format=identity \
    radio-mac=6C:3B:6B:18:A7:F6 slave-configurations=GuestWifi
add action=create-dynamic-enabled comment="Dining Room 2Ghz radio" \
    master-configuration="Skuber 2G auto channel" name-format=identity \
    radio-mac=6C:3B:6B:18:A7:F7 slave-configurations=GuestWifi

Re: hAP AC + capsman performance

Posted: Wed Apr 14, 2021 12:05 am
by en1gm4
For the record, the ethernet interfaces on both devices report no errors (I checked as there is a PoE injector powering the hAP)
Testing between the two routers using btest via winbox gives over 500mbit before hitting 100% CPU
There is only one ethernet switch joining them and everything else on the network seems to be working fine

Re: hAP AC + capsman performance

Posted: Wed Apr 14, 2021 12:34 am
by bpwl

-30: that is almost dead, you probably wanted +30 dBm

17 dBm: common practice in 2.4 GHz band to reduce the power in the hope to reduce the interference with others, but this not done that often in 5 GHz as the walls tend to attenuate the signal more. Lower TX-power is also related to lower interface rates in many cases, but you are very close for the test, and there lower power is OK.
The antenna gain for the hAP ac 5GHz is 2dBi. Max power could be 25 dBm for 5500MHz/Ceee (5490-5570MHz)

[admin@MikroTik] > interface wireless info country-info "united kingdom"
  ranges: 5170-5250/a,an20,an40,ac20,ac40,ac80,ac160,ac80+80(20dBm)/passive,indoor
Expect 400Mbps as a max unidirectional, and 180-200Mbps bidirectional with "866Mbps-80MHz/2S/SGI" interface rate in wireless registration. I don't think your client device picks up the extra 80MHz extended bandwidth (80+80MHz ?)

Re: hAP AC + capsman performance

Posted: Wed Apr 14, 2021 4:20 pm
by en1gm4
I have no idea how the -30dBm got in there.. and it was not showing up afaik in winbox .. i should have thought to look in the config properly.. d'oh. It was not running that for any of my testing though.
per the 17dBm: it's in a house and i have three mikotik AP's on the ground floor covering two floors as the signal travels fine vertically but horizontally there are brick walls to content with. Normally I aim for a lowest signal of -65dBM wherever we generally use devices which seems to work quite well (much better than a single AP blasting out at max like most homes...).
Outside i reduce the signal so that it is just usable at the edge of the property.

5GHz is attenuated much more but I would still use the same approach

I had it on full power for my speed tests though and have reverted to full power again .. sitting near the AP (mounted at 3m up the wall) the signal seems to vary a great deal.... from -35dBM to -47dbM and I've only moved a couple of meters. It stays at 866/866 though (on channel 161 80MHz).. as indicated by Ubiquity's Wifman app.Perhaps that is a function of 5Ghz reflections etc?

iperf remains 151Mbps upload to wired PC and 184Mbps down. (I've read that Mikrotiks have that assymetry with download being faster... seems strange but at least is is in the right direction)

Re: hAP AC + capsman performance

Posted: Wed Apr 14, 2021 6:09 pm
by bpwl
OK. TX-power is no issue here, it was just that odd -30dBm.

But you have more AP's and clients around. So using "Freq usage" and "Snooper" is a must to find out what is near your test devices.

Beware that a low signal level is enough for co-channel interference avoidance (mutex wait for airtime). https://networkengineering.stackexchang ... terference . Having a signal strength of -45 dBm will not eliminate a wait for a -85 dBm transmission.

You might consider using the advanced parameter "Adaptive noise immunity" to reduce this effect somewhat.

Re: hAP AC + capsman performance

Posted: Wed Apr 14, 2021 11:30 pm
by en1gm4
I'll do a little more digging into the use of spectrum around here. The hAP picks up more than my phone does although all at low levels in 5GHz. When I tested, none of my own devices were in active use (4 possible)but that doesn't preclude possible background traffic... although I would have expected at least one higher speed result if it was possible.

It doesn't look like Freq usage and snooper are available in capsman ( or am i missing it somewhere?). I'll have to turn off cap for the 5Ghz interface on the hAP and do it from there presumably.

Given best case clear spectrum, what is the highest likely speed from a single hAP using 2 streams? (rule of thumb 8666/3 would still be 289.. and I'm not close)

This thread mentions getting 400 on an Intel 2x2 on hAP AC and 530 3x3 ( but not on capsman)
Someone mentions getting only 100-130 running capsman but 500 using local forwarding? (although it is not clear if it was HAP AC and 2x2 (when i turned on local forwarding I saw no improvement but it would seem I must have done something wrong so will try again)

I know RF is a black art but it would be nice to know what might be possible given few impediments like co-channel interference etc. Given my hAP CPU is not the bottleneck I assume it must still be in the wireless side ...

Re: hAP AC + capsman performance

Posted: Thu Apr 15, 2021 12:57 am
by bpwl
You should get 360Mbps, or even 400Mbps unidirectional in UDP. (=without TCP ack answers, and no TCP congestion control)
My test was even with devices on different floors. No CAPsMAN used. (Should be the same as with local offloading. However I have seen default settings by CAPsMAN like A-MSDU=2048 in some configs.) I'm just not a CAPsMAN user, still tuning every single AP on its own.

And yes other brands (with more auto-adjust algorithms and bigger buffers get up to 500-590 Mbps.This requires a very large A-MPDU and zero, even idling, AP's around).

"This single client streaming speed test is like testing a SUV on a formula 1 circuit to verify how well it performs. Real life conditions don't match the test conditions."
CAPsMAN is not designed for "single client - single AP" racing.

Re: hAP AC + capsman performance

Posted: Thu Apr 15, 2021 1:36 pm
by en1gm4
so far, having a look at the spectrum and finding the clearest spot (virtually nothing on it according to freq scan), then using that (5180/80 Ceee) has not improved things at all

adaptive noise immunity has not improved things at least on 5180 (which makes sense if there is no noise; i didn't check impact of CPU)

understood re single client testing and mikrotiks target market is generally not homes... but there are lots of installs in homes and small offices where there will only be 1 active user at a time with only a handful of background devices , so it seems a reasonable target for one datapoint (but hardly indicative of real world busy networks certainly ;)

will keep digging... and maybe do a proper non capsman set of tests and see what I can get

any other hAP AC + capsman users with some experience/numbers out there?

Re: hAP AC + capsman performance

Posted: Thu Apr 15, 2021 3:58 pm
by en1gm4
so... perhaps it is not my lacking wifi-fu (or not only that) but a greater ignorance?

I did some more reading and tried using multiple parallel streams in iperf. (not something i have used much until now)
.. and, like magic, I'm seeing impressive throughput... up to nearly 500Mbit with 10 streams. (cpu on hAP at 75-80 percent. CCQ in mid/high 90s)
(on my other 2.4G AP's i have seen mid to high 80Mbit ... for a 144 SYN :)

so now I have questions about why... is it that TCP over wireless really isn't good? do i have retransmission issues? (and where/how would I find them)
am i missing something blindingly obvious?
Does this mean that single tcp stream windows file transfers over wifi will always be similarly limited?
this rabbit hole just gets deeper and deeper

on the plus side it is good to see the radio side doing well.... and for multiple clients things should be ok presumably (i.e a more normal multi device/user environment)
at least I know a little more about wifi tuning ;) ;)

Re: hAP AC + capsman performance

Posted: Thu Apr 15, 2021 5:12 pm
by bpwl
Some more reading .... viewtopic.php?f=2&t=174113#p851770

But here you have no MSS/MTU nor interface speed mismatch. Still TCP congestion control plays a major role.

That rabbit hole is very deep ... things we can do, things Mikrotik can do, but also Linux and even the TCP protocols need adjustments .... if you like reading try this one
from OpenWRT. ... lJBW4/edit

Re: hAP AC + capsman performance

Posted: Mon Apr 19, 2021 1:14 pm
by en1gm4
thanks for this. I did indeed do the reading... and I'm surprised how this is not a more exact science with the money and time spent over the decades... although I guess getting people to agree on anything like "truth" (such as it exists) is not easy. HTTP/3 looks interesting per TCP window etc.

for the benefit of others reading this thread, I found that "Receive Window auto tuning" was, for some strange reason, disabled on my desktop. This was the machine that was used for file xfer and iperf tests with a wired gigabit connection so was giving a misleading view.
Enabling it to default settings made a big improvement in my case.

Currently I have reverted to 40Mhz (PHY of 400) as that meets my needs and avoids interference to some degree, but even with that I have tested single TCP streams to 200Mbit+ and multiple stream to 300 . This is peak and, of course, it varies substantially, but it means things are working well in the RF department for the most part using Capsman and a hAP AC,
Similar improvement in 2.4G but more variable presumably due to interference from more devices and crowded spectrum.