Most of my AP's are based on QCA ath10k (various chipsets, i.e. 2x Archer c2600, AVM 4040, Netgear EX6150v2, 2x Compex WLE900VX in APU2E4).
As I have gigabit internet, I want to be able to get the full wifi speed cumulated. I was testing this via speediest.net and via iperf3 (10GBit/s servers), i.e. running speed tests at the same time.
v(SSID1,t0) + v(SSID2,t0) = 1 GBit/s
whereas SSID1 and SSID2 are from the same AP (having two distinct wifi cards or a single wifi soc with master and vap) at the same time t0.
This works nicely, if there are two (or more) independent AP's (two devices, not SW ap/vap)
But it does not work on a single AP, here two cases
a) APU2E4 with two distinct Compex WLE900VX (ath10k, qca9880, each having a single SSID on 5GHz, different channels. btw with a single WLE900VX you can achieve around 560 MBit/s download and over 600 MBit/s upload wifi speed with an iPad 5 (every time !).
b) Archer c2600, running a master and a vap (i.e. on the same channel...)
Observation for both a) & b): As soon as I start the wifi speed test (iperf3 or speediest.net), one of the clients drops significantly in speed, the other one is then much higher, -> and in sum, the wifi speed corresponds to the speed of a single client.
For case 2a) where there are two distinct wifi cards built-in I would have expected these work independantly of each other (each wifi card being on a separate VLAN, separate ethernet port).
(If I connect each lan (vlan resp.) on a separate AP I get the cumulated 1 Gbit/s speed as stated above)
Is there a possibility to make it run as I would expect or is this an OpenWRT limitation ?
are you aiming for a controlled test? ( accuracy )? why do you make no mention of CPU/interrupts? are you confusing testing wifi with more variables? if so, why have you not included control tests bypassing wifi entirely?
Thanks a lot for replying and for your questions.
I'm not a professional in this area. I hope you can help me whether I measured incorrectly or where I have a bug in my settings.
Maybe my topology helps.
My main router is an old i5-3470, 4GB RAM, Samsung SSD, Dell Optiplex 7010 with 3x i210-T1. On that one I have installed irqbalance and acc. to syslog it seems to work. The i210-T1 have 4 lanes each (acc. to syslog and specs).
The main router is connected with a central switch (Cisco SG300-10), which is again connected to "slave" switches (SG300-10) via LACP, all switches in layer 2 mode.
If I test ethernet only at any switch (eg PC) I get typically 940 MBit/s, as expected.
If I test ethernet only at any switch with two clients (Debian or Ubuntu PCs) I get cumulated also 940 MBit/s, independent if I use the same public 10GBit/s server or different ones (bouygues.fr are the closest to my home).
At the end of one "slave" switch I go out with LAN and a VLAN to an AP (the one I stated in my first post) or alternatively two AP's (one connected to LAN and one connected to VLAN).
in case of the Archer c2600 (VLAN) I installed irqbalance which seems to work.
For testing I added a second AP (avm 4040) for the LAN without irqbalance, set on a different channel on 5GHz than the Archer c2600. Cumulated (same time) I get around 780ish MBit/s. If I use the Archer c2600 alone for both LAN and VLAN (testing at the same time), cumulated I get only 480ish MBit/s, with one client (LAN or VLAN) going very slow and the other one fast.
Considering myself as a layman, I'm stuck at the moment with
a) understanding what I could have made wrong at testing
b) understanding how OpenWrt works and how / if you could find a solution to use only one AP for both LAN and VLAN and get out the most (especially if I could use the APU2E4 with two wifi cards inside).
Thanks for your question and your help in advance.
Maybe it‘s easier to express in a simple formula (I‘ve also adapted my first post)
v(SSID1,t0) + v(SSID2,t0) = 1 GBit/s
whereas v is bitrate (eg in in GBit/s), SSID1 and SSID2 are from the same AP (having two distinct wifi cards or a single wifi soc with master and vap) at the same time t0.
Is my goal clearer now ?
Can you help me to understand if this is achievable or if there is a limitation (by OpenWrt routing or chipsets or wifi drivers) ?
Hmm... I can follow your explained observations and reasoning.
But i do not understand what should be wrong with your explained experiments and what do you think is an OpenWrt limitation.
I‘m interested if it‘s possible to achieve 1 GBit/s in the case where I have two distinct wifi cards on a single AP (the APU2E4 with 2 x Compex WLE900VX, each is a mini PCIe card)
i.e
v1(SSID1,t1)=565 MBit/s
v2(SSID2,t2)=565 MBit/s
but
v1+v2=500+65 MBit/s @ t0 ?? (the values may vary but the sum is more or less the same)
I would understand if it was in a single wifi card, but it‘s two distinct wifi cards.
So it looks like OpenWrt is ignoring that it‘s two cards ? Why, what is the reason ?
btw, the APU2E4 can easily route 1 GBit/s with OpenWrt (4 cores, 4 GB ECC RAM)
Yes exactly, two iPads, each on a SSID,
each SSID on a different channel (36, 80 MHz and 52, 80 MHz) as it‘s two distinct wifi cards (mini PCIe) on a single AP (the APU2E4 with 4 cores and 4GB ECC RAM, you can put two mini PCIe wifi cards in it, plus SSD, plus HDD if you need, and a few USB ports if you need to plug in more stuff https://www.pcengines.ch/apu2e4.htm)
So I‘d expect the wifi cards work independently of each other. But it seems OpenWrt handles them as one, for whatever reason I don’t understand why ?
With route you mean to route different subnets over different ethernet cards, right?
802.11 (wireless lan) != 802.3 (ethernet)
So you can not assume that routing from a ethernet adapter to a wireless adapter has the same reqirements.
This two protocols are very different and transmitting frames between this two protocols needs some cpu power.
Have you ever checked the system utilization on such a iperf3 benchmark?
Not sure if I understand.
The APU2E4 has 3 NICs (3 x i210AT, each has several lanes). On the first i210 I put LAN, on the second the VLAN. Each NIC has a wifi card counterpart.
Do you think, the ethernet to wifi bridges could be a limitation ?
Thanks for your reply.
I tried also channel 36 and 100 on the other, same result. Do you think that the broadcasted signals are detected (assuming in the low dBm range still a signal) by each card from the other band and therefore jam each other (although each card is connected to 3 antennas, but yes cards are a few mm from each other) ?
Best case 4x4 AP connecting with 4x4 client - you’ll saturate the access point ethernet- see the smallnetbuilder article with two true 4x4 devices connecting with each other in an optimized environment:
**** take in mind that from this article it is one client connected to one access point****
Multiple cards = multiple clients
940mbps / 2 clients = best case 470mbps each client (CPU limitations and interference will prohibit you from reaching this).
Wifi has overhead and interference issues, as more devices start competing for bandwidth on the same channel it gets a little more crazy and you don’t get the bandwidth you were hoping for:
I made a poor man's repeater out of my Archer C7 by replacing the PCIe 5 Ghz card with another Atheros 2,4 Ghz one so that I had onboard 2,4 Ghz in parallel. Even though it worked and I used an external Antenna meters away, performance significantly degraded as soon as I used both in parallel on different channels. I assume that would be also the case if you just try to use two 5 Ghz cards. Either go as suggested for 4x4 mimo or 160 Mhz if you need that much bandwidth.
I've made now the same experience as JevermeisteR and mk24 explained.
Result: The cards or rather strong signals are jamming each other, even if you put channels 36 and 116 (or 132), as the farthest possible (in my country).
Only solution: Put the two emitters (AP's) approx. 2 meters from each other with least spatial beam intersection.
It was easy to measure (using iperf3 on notebooks).
a) two identical AP's, two different SSID's, each set with 5 GHz, 80 MHz BW, but different channels (sub-configs: main frequencies close and far from each other, i.e. extreme cases)
b) two notebooks, first notebook wireless connected to first AP, second nb to second AP, both notebooks with iperf3 (unixoid OS's)
c) distance between AP's, whereas antennas were all time parallel to each other (for the spatial beam direction, short antennas meaning large emission angle)
-> started with "0m", 0.25m increments up to 2 m (my current cable was too short to go farther, practical limitations...)
Observations
distance < 1m between AP's
-> strong jamming, cumulated speed low < 50 % (or maybe as fast as with a single AP)
1 m < distance < 1.5 m
-> low to moderate jamming, 65% < cumulated speed < 80%
1.5 m < distance < 2 m
-> low jamming, 80% < cumulated speed < 90%
distance ~ 10 m (this was another AP, not identical)
-> cumulated speed ~ 100% (almost 940 MBit/s)
(- reference measurements with both notebooks connected via ethernet on the respective AP, gave always cumulated (i.e. running iperf3 in parallel on both nb's) ~ 940 MBit/s)
In my case it was not the CPU (thx for the hint juppin and Gnufsh).
Thanks again !
This means I'd need two AP's with a certain distance or maybe a 4x4 powered AP (thx ACwifidude and JevermeisteR)
I will dump a bunch of links on you, since i have played with Archer A7 v5 (same hardware as C7 v5) to understand it's bandwidth limitations. Being a single CPU router, CPU irq handling is the limiting factor.
Test with alternative "ct-htt" firmware, ie:
opkg remove wpad-basic ath10k-firmware-qca988x-ct &&
opkg install wpad-openssl ath10k-firmware-qca988x-ct-htt
?
?