Catalyst 9130axi throughput issue

[Radu96069]

Hello,

I think there is something wrong with the Cisco 9130 EWC I have because no matter I tried I never got more then a bit over 600 Mbps on a connection that can do 944 (1Gb switch) even though on Ruckus R710 I get more than 817 Mbps form one 3x3 connection (iMAc).
I removed all Qos and AVC settings but there was no effect.
There is only one SSID and no interference on the channel.

For the 2x2, 5Ghz, 80Mhz the throughput is somehow what it should be (even though Ruckus still performs better) and therefore I didn’t observe until now that there is a potential issue. I realized only when I added the 3x3 iMac and the wifi 6 160Mhz Dell laptop.

Does anyone have any idea why this happens?

Devices used for testing.

Cisco 9130axi: 4 antenna radio (4:4) and 8 antenna (8*8) 5Ghz ax radio setup (same results)

Ruckus R710: 4 antenna radio (4:3) 5Ghz ac radio

Cisco ISR 1100 – both APs and Iperf3 server are connected to it

60w poe injector

Dell laptop wifi 6 2x2 160 Mhz – 2402/2162 link on Cisco, 866 on Ruckus

Iphone 11 – wifi 6 – 2x2 80Mhz – 866 link

Imac -wifi 5 – 3x3 80 Mhz 1300 Mbps link

Samsung Galaxy Tab S7 wifi 6 -2x2 2.4 Ghz, 344 Mbps link

Cable, switch and iperf server tasted with Ethernet first: results 944 Mbps for both uploads and downloads (cable is 6a but I do not have any multi gig switch yet)

Iperf command: iperf3 -c <ip address> -P4 -R -t10

Testing results

• Iphone 11
  • Cisco – max 606 Mbps (
  • Ruckus – max 668 Mbps
• Dell laptop
  • Cisco – max 582 Mbps (this is strange on a 2162 link)
  • Ruckus – max 502 Mbps
• iMac
  • Cisco – max 571 Mbps
  • Rukus - max 817 Mbps
• Galaxy Tab
  • Cisco 2,4 – max 228 Mbps (it didn’t connect on 5ghz no matter how many time I disabled and enabled the wifi)
  • Ruckus 5Ghz – 645 Mbps
• Galaxy Tab on 2.4 Ghz + iMac on 5Ghz for 5 minutes
  • Cisco – max 510 (aprox) total throughput
• Iphone 11 on 5Ghz + iMac on 5Ghz for 5 minutes
  • Cisco – max 530 (aprox) total throughput
    
    

On an AP with a total theoretical throughput of more than 5 Gbps to get a max of 0.6 Gbps in almost ideal conditions it is not ok. Especially when with the exact same conditions, I get 0.8 from a single connection to Ruckus R710.

Another thing that I observed but didn’t have time to properly test is that if I run iperf on 2 devices on the same time that are connected using different radios (either 2.4 and 5 or two 5Ghz) the connection o one device crushes and I need to restart it. It happened several times during the tests. If the devices are connected to the same radio (I tried with 3 connections at the same time) everything works fine.

Thanks!

[Leo Laohoo]

---

@Radu96069 wrote:

60w poe injector

---

With 60w injector, does the radio report "full power"?  

Kindly provide the firmware used. 

When doing all these tests, what is the neighboring spectrum like?  Are these tests conducted inside a "sealed" room?

---

@Radu96069 wrote:

 I do not have any multi gig switch yet

---

Noted.

Mike Albano has a very comprehensive list (LINK) of every wireless clients and what it can do.  
Important to me are the supported channels, whether they are "really" 802.11ac/ax or just "wannabes".  

[Radu96069]

Hi Leo,

Yes, the radio reports full power. I also tested with a Poe+ switch (still full power and the switch reports a 13W power used on the AP port)
In both cases the usb is disabled.

I use 17.6.2.43 with Boot version 1.1.24.

There is no interference reported on the 5ghz channel (36). In fact there is none on any 5Ghz channel as the APs are in a semi isolated house not in a building. The ap shows 1% channel utilization and 9% on chnnel 1 (2.4Ghz) as there are 3 APs 'visible' from neighboring houses (-80 -90 dBm).
And still the environment is the same for both APs used for testing.
Also the device capabilities and connections I've got from the devices but also from the AP. The links reported by the AP are the same with the ones reported by the devices.

Thanks!

[Arshad Safrulla]

1. Shutdown the 2.4 radio in the WLC and do not enable dual 5GHz. (Default will be only 1 5GHz radio). If you want to test these then create new test cases specifically targeting these.

2. 9130 with usb port enabled you need UPOE, either use a UPOE rated POE adapter/ UPOE capable switch to power the AP or disable the USB port.

3. Ruckus 710 is AC and 9130 is AX. So there is a dependency on device drivers here. Use the latest drivers and latest OS patches. Windows/Apple all known to have bugs which impacts the speed/ throughput due to bad drivers. 

4. Running EWC on top of the AP will add certain load in to the AP, even though it's not documented I assume that it will have an impact on the AP performance., so if I were to test I would fire up a 9800-CL in any supported hypervisor and use Flex connect mode (local mode requires dedicated compute resources) or use Local mode with any physical box. 

5. If your aim is testing the speed only, then disable authentication and make the SSID open and test. If you want to test with Auth enabled do not enable any WPA3, PMF etc.

6. Also record the MCS rates client negotiated each time before you start the test, also check from the AP side how it sees the client. 

[Radu96069]

Hello Arshad,

1) I've done that, same issue. What intrigues me is that the AP behaves like is throttled at 600 Mbps and does take advantage of the 3rd stream available on the iMac (as Ruckus does). I've restarted the 2.4 radio and I created a second SSID only wiht 2.4 radio. I've got on that one 150 Mbps with the 3:3 iMac (good enough for a 3:3) and 200 with the iPhone (ac vs ax but on a 2:2 setup).
When I run iperf on a single machine in 5HGhz I get a ~600 Mbps max speed. If I start the iperf also on the new 2.4 ssid the speed decreases on the 5ghz to ~450 Mbps. This is why it looks like the AP is throttled. Whatever I do it cannot go faster than 600 Mbps.

2) the USB is disabled. I've tested with 60w power injector and with a poe+ (30w) switch. The power consumption is less than 15W even when I run 3 iperf test in parallel for 5 minutes. The AP shows full power so I do not think there is a power issue.
3) Everything is up to date. Nothing I can do here. And there are different devices with different chips and drivers. All have issues?
4) Not something I can do (not enough time and knowledge to start doing it). Also the processor is at 15% utilization when I run the tests. There is only one AP so the controller overhead shouldn't have such a big impact on performance. The R710 also runs the Unleashed controller but still operates in the normal parameters.
5) I used wpa2 personal for the test on both APS. R710 allow a 800 Mbps connection on iMac, 9130 doesn't. Also the iphone throughput is better on R710. (just to make sure is clear, I do not compare Ruckus with Cisco. It is just the other AP that I have installed in my small office. It is an overkill from cost perspective but I need high stability and high throughput for the few machines that are used and I cannot run cables
6) The rates are reported by the AP. Normally, the 6Mbps rate is reported as there is not much traffic and this is the lowest available rate. When I start the test the AP reports what I mentioned in the initial message.

Thanks!

[Arshad Safrulla]

I would suggest you to downgrade to 17.3.4c or upgrade to 17.6.3

[Radu96069]

Thanks, Arshad.
I will stick with 17.6.2 for the moment because I do not have anymore time to do the work of Cisco QA people and I'll wait for the TAC recommendations (for the moment they recommend 17.6.2; March 31) to avoid as much as possible problems with my network. I will redo some of the test when I have the new software version deployed and I will update the post.

[Leo Laohoo]

I think 17.6.3 will be out by the end of this week.

[JPavonM]

17.6.3 was released a few days ago:

https://software.cisco.com/download/home/286321396/type/282046477/release/Bengaluru-17.6.3

[Leo Laohoo]

---

@JPavonM wrote:

17.6.3 was released a few days ago:

---

Oooops.  I got my release dates crossed.  

Thanks for the correction.

[Radu96069]

Update!

Some additional test results (some quite unexpected) and some open questions to hopefully be answered by those who know more about the topic.

C9130axi real throughput on 2.4Ghz is 72% of the theoretical one on all devices (including the 3:3 iMac and a 1:1 Raspberry Pi ). So, I found no issues on this radio.

In the 5Ghz radio (4 or 8 antennas) the real throughput reaches 65% of the theoretical one for all 2:2, 80 Mhz channel devices. The 3:3 iMac has the throughput of a 2:2.

Dell on 160 Mhz channel is a lost cause for the moment because I couldn’t make it work but as I do not have any other wifi6 ap or device that can do 160 Mhz I cannot tell if it is a device or ap problem. Both report proper link but the throughput is equal to an 80 Mhz channel.

I’ve tested more the 2 radios scenario using the same or different SSID.

2.4 and 5ghz different ssid

• iMac on 2.4, iphone 11 on 5Ghz max throughput less than 600 Mbps, equal to iphone 11 alone. So, the throughput decreases even though 2 different radios are used. Not ok from my perspective.
• Iphone on 2.4, iMac or an wifi 5 Dell laptop on 5Ghz, finally the total throughput passes 700 Mbs (~550 + ~149)

Two 5ghz Radios (channel 36 80 Mhz, channel 100 80 Mhz) same SSID

• I manage to make only the iPhone 11 stick to the second radio and for a brief period the old wifi 5 Dell. Unfortunately, on all scenarios I managed to test the total maximum throughput was less than 550 Mbps. I was expecting to have 450 (iPhone on second radio) + 550 Dell/iMac on the first one. But they behave as running on the same radio.
  Any ideas why? This is important as I would better install 2 2x2 aps instead of the highly expensive C9130 and it makes all micro-macro cells capability a marketing lie.
• Also, I observed that the connections were quite unstable as the AP was probably trying to optimize the channel utilization and “moved” the devices from one radio to another (2.4, and the two 5 GHz quite a lot). That is not a scenario I want as people spend many hours per day in video conferences.

Cisco ios 17.6.2 has some bugs as it takes minutes to delete the disconnected devices from the list (command line and Gui), the MCS reported by the controller are not in line with the rates reported by the AP and the iperf reported throughput  (show wireless client summary details vs show client summary). For example, the rates were 1300 and the MCS was 15 or 16. Only once I managed to see a Mcs 21 reported. MCS 15 is also strange next to a 3/3 NSS because based on Cisco MCS table M15 is a 2/2. Most of the time for the wifi 6 devices the MCS as 9 but the reported rates and the actual throughput were MCS 11.
Channel utilization is also misleading or not real time. I’ve got 62% channel utilization on radio 2 even though there was no device on that radio (no interferences either) and TX and RX channels showed 0 utilization.

And a final thought about something I realized reading the products specifications.
None of the Apple products is Mu-MIMO capable so in this case 8 antennas radio AP is as useful as 2 antenna radio (I did a quick test on an 2x2 Ubiquiti outdoor AP and I’ve got similar throughput with 3 devices. Hopefully with 30 is a different story).
If your network is Apple dominant have a look into this. You might save some money and have a better wifi network setup and user experience). In my setup an 8 antenna MU-MIMo ap would have been perfect based on the devices number and locations but unfortunately it seems to be useless as most of the devices are Apple. Hopefully OFDMA capability helps but I couldn't see any improvements testing with the wifi 6 dell and the iphone 11. There are different scenarios but in my case with a few users that move and work on huge files, throughput is most important. Latency comes next and it is anyway good enough for Zoom calls.   
I was not able to test the MU-MIMO capability and throughput as I have only one device that theoretically has that capability (the wifi 6 Dell).

Any clarifications from Wi-Fi experts would be much appreciated.

Thank you!

[JPavonM]

@Radu96069 have you tried updateing Dell's wireless chipset drivers with latest Intel's ones? Tehre are many fixes that maybe of interest for your tests.

https://www.intel.com/content/www/us/en/download/19351/windows-10-and-windows-11-wi-fi-drivers-for-intel-wireless-adapters.html

[Radu96069]

@JPavonM, Hi

I've checked before the test and it had the last driver. But this week I will update it to Windows 11 and try again.

Thanks!

[Radu96069]

@JPavonM 
Hi,

I updated the driver with the one on Intel website.
Unfortunately still not getting the 160Mhz throughput.

I'm wondering if there is anyone that managed to use the C9130 with a 160Mhz channel that can provide some details.

[Jan89]

Hi!

I thank you for publishing this tests. Recently, I was considering buying 9130axi to use with EWC.

I am currently using both Cisco AP3802 and AP4800 (running Mobility Express which is the equivalent of EWC) with MacBook Pro 15 (2017 model) and Macbook Pro 16 (2019).

Both MacBooks has 3x3 wifi cards. What is interestingly older MacBook is achieving better results in iperf test and speed tests.

My results:

MacBook Pro 15 (2017 model):

iperf: over 900Mb/s in one TCP stream (I saw the peak results about 930Mb/s). This is on 1300Mb/s connection rate.

speedtest.net: about 870Mb/s DL/UL speeds.

MacBook Pro 16 (2019 model):

iperf and speedtest are about 680Mb/s max also on 1300Mb/s connection rate.

But it looks like a MacBook 16 problem - people complains about slow wifi speeds on this machines

https://forums.macrumors.com/threads/2019-macbook-pro-16-slower-wi-fi-speed-compared-to-2015-mbp.2215930/

iPhone Xr:

max 640Mb/s on speedtests which is good result for 866Mb/s connection.

I also don't think that running EWC will add such load to 9130AXI to slow down this AP which has quad core CPU.

AP 3802 and 4800 has dual core CPU and can achieve >900Mb/s with 3x3 clients.