'ax hardware 'speed limit'

RouterOS Wireless Networking
1

Hi all. I’m having trouble getting a UDP speed test between a mANT 'ax and SXTsq 'ax beyond about 475Mbps.

I can get 470Mbps on 40Mhz. However, if I bump that up to 80 Mhz I get slightly less.

AP on 5180/ax Ce is 470, Ceee is 415.

These are forced 5Ghz AX, WPA-PSK, cipher is GCMP256. Speed test shows CPU load on both at around 50% so seems like there’s headroom there.

I’m just not sure what I’m missing here trying to get more speed.

20Mhz modulation is at 286.8 and actual speed is ~210, 20/40 jumps to ~573, 20/40/80 to ~960, and …160 is 1297.

I do get brief spikes up to about 620mbps on the larger channels but they’re just a few seconds and then back down under 500Mbps.

This unii1 band is clean so this isn’t just some noise issue.

Worth noting that on higher channel sized CPU runs up to as much as 80% but without more throughput.

2

I don’t have much experience with MT wireless but just some sanity check questions. When you watch the CPU, do you watch each core utilization? What sort of UDP test you are running, have you tried multiple streams? What about TCP tests?

3

The numbers sound similar to what I was able to achieve at 80/160mhz with a similarly specced wap ax.
(same CPU + 5ghz as sxt)

CPU gets a lot more usage though at 80% but that includes vlan tagging. (which also runs on the cpu cause of the wifi)

To be fair i had “normal” wireless clients connected. (Phone and Laptop)
And UNII2 isn’t especially clear here. And my intel wireless doesnt support UNII3

It got me kinda disappointed with the wap ax cause an ax3 gets more throughput at only 80mhz.

4

I can’t reliably get any faster, If I run a few hundred connections through I can get a bit over 500Mbps in 80, but with <100 connections TCP 40 and 80 are basically identical. It’s like there’s an invisible speed limit somewhere in the system.

5

What are signal strengths reported by both sides? Beware that doubling the bandwidth (from 40MHz to 80MHz) will decrease signal strength by 3dB. And if signal strengths are not great (read: above -60dBm or so). this can lead to loss of speed due to required Rx sensitivity (both SXT sq 5 ax and mANTBox ax require -67dBm for highest speeds and Tx power capability at those speeds are not stellar either: 16dBm and 20dBm respectively).

So in conditions where Rx signal strength doesn’t allow for comfortable communication at highest rates, doubling the band width doesn’t improve anything, more likely it causes less stable communication and slight rate drop.

6

this is in my lab so ‘whatever I want’. I tested at anything from -55 of both sides to -30 on both sides. 10 feet apart facing each other, max output set to 0, -40 signal on both sides. live iperf flowing across the connection and I just can’t get it to go over those speeds.

note, there have been zero changes to queues, the radios are bridged and fast forward is enabled, there are zero firewall rules, if set MTU up to 1560 just to make sure there’s no fragmenting, I’ve tested 7.18.2 and now I’m on 7.19rc1, and both devices are on wifi-qcom package.

The radio is modulating at 2.1Gbps and I can barely get 400Mbps across it on clean air with perfect signal.

7

maybe not, this devices have 2 core CPU, 50% of cpu usage can mean you have maxed one cpu core at 100%, is a periodic ocurrence in MikroTik devices, just hope in the future this situation improves or maybe a different kind of testing environment can allow to split load on both cpu cores

in the past i had the same situation on audience (RBD25G-5HPacQD2HPnD) on the third radio cannot exploit the 160mhz wide channel potential because one cpu core maxed up

8

Also, if I connect my iPhone16 I can get a burst up around 850Mbps, then that sinks down to under 600M. So faster, but still no where near where I’d expect a 160Mhz WiFi6 channel to be.

I’ve also tested the 2.4AX to iphone and modulation says 229Mbps, I can get around 84Mbps.

These tests from a unifi U7 Pro yield twice these speeds, and from an eero 7 even more. Similarly, the unifi ux7 immediately hits 950Mbps on iperf which lines up with it’s 1G port limit.

Either I’ve missing some key piece on how to extract performance from these or the platform needs work.

I got these hoping to have an alternative to Cambium’s wifi6 gear but I can’t even get these to perform on the bench…

9

I have not, looking at the CPU profile each individual core is only at 50% or so also.

10

more testing. this is really clearly a client side issue. I connect it to a different wifi7 AP (unifi u7 pro) and get same speeds. iPhone to that is over 800Mbps consistently.

11

I just swapped the sxtsq-5ax for another mANT-5ax in client mode. I’m hitting a similar speed limit here. CPU cores are well under 70%, nothing in the way as far as I can tell, modulations clear up at 2.4Gbps, can’t break 500Mbps mikrotik-to-mikrotik. same mANT-5ax as a client to u7pro, just cannot break 500Mbps.

If this is a CPU limitation, the use is hidden, resources or profiler can’t see it.

since the mANT has 2ax I decided to test that as well. it’s also very slow. 18-45Mbps on 20Mhz where an iphone is doing 100Mbps. this is also to the unifi u7pro which has 2ax and 5ax, and the radio is showing 2Ghz AX or 5Ghz AX.

12

There were people on some other topics talking about changing the interface queue type from hardware-only to wireless-default to increase throughput.

Your 2.4ghz seems indeed very slow.
I was still able to push my line speed through the wap ax (although that’s just 100mbits) but I didn’t test more than that

13

the wireless interfaces are already set to wireless-default.

14

I was referring to this topic where they mention “wireless-default” on ether1 but I didn’t get to testing that yet myself.

http://forum.mikrotik.com/t/slow-ax-wifi-on-wap-ax/182093/15

15

I’ve tried all sort of queue types. fq_codel, sfq, hardware. There’s CPU left over on all of them but throughput doesn’t go up. Cake works as well but only up to about 400M before cake is the CPU limiter.

Seems I’m not alone here with the poor performance on the CPE side. I just wish there was some indicator for why. The usual suspect of CPU limitations really doesn’t appear to be the issue.

16

btw i just did a couple more tests and i got the following results against the wap ax with an xiaomi 12 pro as iperf server

All Single Connection. All 160mhz for 5ghz and 20mhz for 2.4. Overlapping with neighbor wifi. Also the wifi signal might be overloading the Phone at -35
Wap AX doing VLAN Tagging for all traffic tested.
Wap ax config is “caps mode” + VLAN Filtering on the bridge.

UDP Iperf:
Server Receive 5ghz:
743 Mbit/s (CPU Core 0 hit 100% CPU and IRQ)

Server Send 5ghz:
229 Mbit/s

Server Receive 2.4ghz:
113 Mbit/s

Server Send 2.4ghz:
97 Mbit/s

TCP Iperf
Server Receive 5ghz:
424 Mbit/s

Server Send 5ghz:
590 Mbit/s

Server Receive 2.4ghz:
100 Mbit/s

Server Send 2.4Ghz:
104 Mbit/s

17

pretty much in-line. There’s a major performance issue here. It’s basically half what’s expected. Super dissapointing.

18

I created a support ticket as i wasnt able to increase speeds in any way.
Even without VLAN Filtering or on 80mhz (didnt bother trying 40mhz)
Single Core CPU load without UDP Receive never goes above 40. Only on UDP Receive 1 Core hits 100% use at 700Mbit/s as stated before.

I’m able to achieve about 2x TCP Receive speed with an “in production” ax3.
814 Mbit/s and that does include VLAN filtering and its overlapping with even more neighbor aps (wap ax just had 1 far overlap)

19

Make sure you are testing to-from PC, not to-from the router itself. And make the PC be the iperf server, not a phone. You might be hitting CPU limits somewhere.

PC with IPERF —> wAP AX ----> PC with IPERF

20

I was 100% sure thats a reply i was gonna get :smiley:

So i did the perfect test setup:

Now i got an:
WAP AX (AP)
hap ax² (Station Bridge)

Both updated to 7.19rc2
Otherwise nothing has changed.


On both ends i have a client connected via 1G ethernet.
AP end is a Windows 11 System.
Station end is a Ubuntu System acting as the Iperf Server for all tests.

Config is basic. Hap ax2 running default - firewall rules - dhcp + station bridge setup.
Wap ax is still running “caps mode” config with a single wifi config specifying nothing more than SSID, PSK and Country (Germany).

Signal is around -45

TCP Iperf:
Server Send: 425 Mbit/s
Server Receive: 248 Mbit/s

UDP Iperf:
Server Send: 743 Mbit/s (Station CPU1 at 100%)
Server Receive: 582 Mbit/s (AP CPU1 at 95%)

For another test i connected the AX2 Station to my “in-production” AX3s wifi (7.19rc1) and moved the Windows 11 System these are the results:

TCP Iperf:
Server Send: 461 Mbit/s
Server Receive: 234 Mbit/s

UDP Iperf:
Server Send: 653 Mbit/s
Server Receive: 303 Mbit/s
Keep in mind the ax3 is in use and also transferring data to another client at the same time.

And with the Windows 11 Client connected via wifi to the wap ax and ubuntu system via ether2:

TCP Iperf:
Server Send: 542 Mbit/s
Server Receive: 711 Mbit/s

UDP Iperf:
Server Send: 767 Mbit/s
Server Receive: 729 Mbit/s

And comparing this to the AX2 as an AP instead:

TCP Iperf:
Server Send: 612 Mbit/s
Server Receive: 482 Mbit/s

UDP Iperf:
Server Send: 795 Mbit/s
Server Receive: 698 Mbit/s
The wap ax had to overlap around 80mhz compared to the ax2 which had no overlap at all.

EDIT: I forgot to mention the overlapping network was quite far away with around -85
Also the Windows 11 Client has a Qualcomm FastConnect 6900