BCM6750 switch function in AX58U - relating to wifi throughput, latency and jitter, microstutter in PCVR.

[munkiemagik]

I've been lurking for months here, finally need to post something. I'll try keep it short (I failed). If something doesnt make sense or needs further clarifying please ask.

It lists on techinfodepot that the GbE switch of the AX58u is not a seperate component it is integrated into the BCM6750 soc. So the BCM6750 is an all-in-one integrated solution for cpu, switch, wifi radio

I've been struggling with the odd microstutter in my wireless PCVR at high framerate (120fps) and high bitrate (500mbps) simracing.
RTX 4090 PC to Quest 3 VR headset. 5GHz band at 160MHz wide in DFS channel.

I've done a lot of testing, diagnosis and elimination in trying to identify source of problem and find a solution to elminate the last bits of microstutter. And the issue (most probably) is not my PC hardware frame generation or the encode/decode chain of such hight bitrate video stream to Quest 3 VR headset, nor a wifi neighbourhood congestion issue. So it must arise from some shortfall in the network transport.

I am going to attempt one last experiment. By using the ISP router (insted of AX58u) to take care of routing for the home but then using a switch for the PC and AX58u, turning the AX58u into a dedicated access point for Quest 3, It was already the access point solely for the Quest 3 but also running routing duties and swtiching for another AP. The bandwidth useage, routing requirement from the other AP is virtually nonexistent.
Am I really going to see any improvement in network transport by turning the AX58u into dedicated AP mode behind a switch? Becaue even though the AX58u CPU is no longer concerned with routing duties, I'm not really avoiding any bottlenecks in the BCM6750 soc as the packets are still travelling along the exact same path inside the AX58u BCM6750 soc as before?

The swtich I have ordered has an 'alleged' 60gbps backplane throughput and forwarding rate of 48million packets per second. But if i belive the microstutter is induced by some bottleneck in the BCM6750 soc then I cant really avoid that pathway bottleneck even by disabling router functionality in the AX58U becasue it is an all-in-one integrated system right?

So if this switch yeilds no discernible change then the solution would be more performant access point hardware?

 

[TheLostSwede]

Well, that router is a low-end model, which has a low-end Cortex-A7 based SoC in it and it's not known as a great router SoC. In fact, the entire platform is somewhat crippled, as the SoC only has support for 16-bit DDR3 RAM, whereas high-end router SoCs use two 16-bit chips (for 32-bit width) of DDR4. Also note that only the 2.4 GHz WiFi radio is integrated, the 5 GHz one is a separate chip.

The WiFi hardware in the Quest 3 is also "limited" to 2x2 for WiFi (as is your router), which means you have limitations on that end as well, even though 160 MHz wide channels are supported. That said, the Quest 3 is using Qualcomm hardware and your router Broadcom, so there can be some issues there too, as across brands, there can be some WiFi related "bugs" that shouldn't be there, but as everyone does their own implementations of the WiFi standard and then add their own secret sauce, this has been ongoing for generations of WiFi hardware by now.

It's very unlikely that it's a switch issue, as Gigabit Ethernet switches haven't had any switching related problems for at least a decade by now. Also, the marketing on the switch you bought is across all ports and the switching IC inside of it, will only switch so much data per port.

You might want to look at getting a more high-end router or AP though, considering what you're using it for, as it could simply be a matter of the SoC not keeping up.

This might be useful too

Re: Meta Quest 3 bitrate and latency for PCVR over Wi-Fi 6e, Air Link, and USB C

My issue seems to be resolved, but a few things changed so I'm not sure what fixed it: 1. I added a *dedicated* router for the Quest, tplink AXE75/AXE5400, as a 6GHz access point. 2. Meta seems to have pushed a small update that improved Airlink performance and turned on a few features, even...

communityforums.atmeta.com

 

[munkiemagik]

Thanks for reading all that and responding @TheLostSwede, really apreciate it..

You make a really good point about GbE switching being so mature now that even an integrated GbE switch shoudnt really have any issues with the useage I am talking about. Max throughput from PC to Quest 3 is showing around 500mbps with my desired use case. In throuhgput tests I have consistently observed the network being capable of stable 970mbps from PC to Quest 3 but of course sustained throughput doesnt really show the lag and jitter which can detrimentally affect VR streaming.

So it really must be a general BCM6750 issue.

Another incredibly valid point you rasied was the Quest3 Wifi hardware/firmware. I tested even higher bitrates (720mbps) using the USBC link cable and it was flawless. So that demonstrated the issue isnt in the encode/decode chain but that testing doesnt eliminate the Q3 wifi hardware itself as a suorce of the microstutter. I can only optimistically hope that the SXR2230P mated to Fastconnect 7800 and all other associated modern components in the Q3 are up to task for modern throughput needs and arent operating on the very edge of their performance envelope at such an early stage of the prdouct lifecycle.

The plan was always to go with an aarch64 board openwrt router (or maybe x86-64, if I decide I want to play around with proxmox and self-hosted cloud storage) paired with a more modern dedicated AP platform. I'm just waiting another month for current ISP contract to be close to penalty free termination so I can switch to FTTP before I go on a big network hardware shopping spree.

I was just hoping there might have been one last hail mary save somewhere before I had to cough up for a setup like; BananaPI R3 as router and 2x Ubitquiti 5/6GHz AP. Its a lot of spend JUST to not have the odd microstuter of the background landscape sweeping across my eyes in corners as I UNgracefully play 'race' around the Nurburgring in VR, practising for when I make the trip with my cousin one day in real life, lol.

Appreciate the Meta forum link you found. will go over that now

 

[Tech9]

So it really must be a general BCM6750 issue.

Disable all Trend Micro bloatware, withdraw the data sharing agreement in Privacy, reboot your router and try again. If still not good - think about TP-Link Archer AXE75 with dedicated 6GHz radio to the VR set. Why this one? Because I know two kids around using it with VR sets and happy. Found later on gaming sites this model is often recommended for exactly this purpose. It's often around $180 on sale and apparently works well. Expensive doesn't necessarily mean better.

 

[munkiemagik]

... the Quest 3 is using Qualcomm hardware and your router Broadcom, so there can be some issues there too, as across brands, there can be some WiFi related "bugs" that shouldn't be there, but as everyone does their own implementations of the WiFi standard and then add their own secret sauce, this has been ongoing for generations of WiFi hardware by now....

I was aware that different manufacturre hardware can require modification of drivers and firmware etc within a system ie using a mediatek cpu with qualcomm or boradcom wifi. but I didnt know that this extends to when packets are already in the air? I always imagined the source 'brand' of the packets would be invisible once in the air. (I dont know anything about network hardware and protocols or the 'construction' of packets as they travel up and down the first two OSI layers, so what headers are attached where and why etc) Do you have some suggested further reading on this you could point me to please just to boraden my understanding, thanks

 

[munkiemagik]

Disable all Trend Micro bloatware, withdraw the data sharing agreement in Privacy, reboot your router and try again. If still not good - think about TP-Link Archer AXE75 with dedicated 6GHz radio to the VR set. Why this one? Because I know two kids around using it with VR sets and happy. Found later on gaming sites this model is often recommended for exactly this purpose. It's often around $180 on sale and apparently works well. Expensive doesn't necessarily mean better.

Its on merlinwrt with absolutely nothing unnecesary running. Only 5GHz radio enabled in DFS channel strictly for the Quest 3, no other wireless devices connected, cant use other non-DFS channels as my wifi neighbourhood is disgustingly congested. My nieghbours have these obnoxious Tenda mesh systems that spew a billion 5Ghz SSIDs all over the regular channels at ridiculous TX power.

I have a seperate AP which runs off the AX58u to the lower two floors for general use. But there is barely any activity on those lower floors when I am up on the third floor, just a printer and a ring doorbell, oh and a lady watching juggling monkey videos on facebook. Dont tell anyone but that second AP runs through powerline as I havent gotten round to doing a cable run, but it doesnt pose any problems as there is no bandwidth need down there save for a living room TV which noone watches.

This is why I really want to deploy 6GHz upstairs when I make the switch to FTTP when my current virginmedia contract ends. But I have had some issues with 6GHz on the Quest 3 and went to reddit to find if others were havng issues, seems since Meta's v64 firmware update there are a lot of people struggling with 6GHz. Ive tested a few 6GHz routers myself and have had unstable erratic performance on v65 firmware, much worse than my AX58u on 5GHz@160 wide.

But if/when that gets sorted its looking like I may end up deciding to order two Unify AP's to mate with an aarch64 openwrt router. If I really need the 6GHz becasue of my neighbours, I'm considering the unify U6 Enterprise (preowned) or U7 Pro depends on whichever I can get cheaper for upstairs Quest3 use and for the lower part of the house the U6+/lite. I suppose I dont really need a second unify AP for the lower floors could just AP mode the AX58U and use that, it will just stand out more.

All this is what started me lurking on SNB and Openwrt and techinfodepot etc the last few weeks. I realised for the price of a TP-Link AXE75/Asus RT-AXE7800 I could build a filogic 830 openwrt router with MT7916 Wifi 6E AP module for much less and have multigig capability, I just have to learn how and all the why's and wherefore's. Which is why lately Im bothering all you nice people in all the forums

 

[Tech9]

my wifi neighbourhood is disgustingly congested. My nieghbours have these obnoxious Tenda mesh systems that spew a billion 5Ghz SSIDs all over the regular channels at ridiculous TX power.

Now many APs you see around you and with what signal strength - it doesn't matter. Tx power can't be higher than your local RF regulations. Applies to you and to your neighbors. What matters is channel available bandwidth. How did you figure your Wi-Fi environment is disgustingly congested?? Many folks do this mistake looking at data from Wi-Fi Analyzer type apps without understanding what it means and select worse Wi-Fi channels as a result.

 

[sfx2000]

The WiFi hardware in the Quest 3 is also "limited" to 2x2 for WiFi (as is your router), which means you have limitations on that end as well, even though 160 MHz wide channels are supported.

Remember that with 160MHz channels, there are WiFI chipsets that will fall back to SISO operation (single stream).

For the Meta Quest unit, I would recommend a dedicated Access Point and SSID just for it. That way, the Wireless connection is not shared with other devices, given that wireless is always a shared medium, but from AP to Client, a single Client on that AP should get all the resources.

 

[munkiemagik]

Now many APs you see around you and with what signal strength - it doesn't matter. Tx power can't be higher than your local RF regulations. Applies to you and to your neighbors. What matters is channel available bandwidth. How did you figure your Wi-Fi environment is disgustingly congested?? Many folks do this mistake looking at data from Wi-Fi Analyzer type apps without understanding what it means and select worse Wi-Fi channels as a result.

HI, thanks for jumping in. I have been wonderig about what it really means to be congested and even with overlapping channels and at what dbm levels things start to get noticeably degraded etc.

This is a row of terraced houses and the houses behind are vry close as well as we literally have no gardens in our par of town.

Curretnly I have 9 5GHz networks around me. 3 of them in the -40 to -50 range, two in the -60 range and all the rest in the -70 to -80 range. And all of them spanning across channels 36-52.

I was probably just beign daft when I talked about Tx power. Its just that sometimes my neighbours 5GHz mesh are showing almost as strong as my own 5GHz when Im right next to the AX58U.

I've never run any Lan bandwidth tests with the AX58u on one of these occupied channels with other neighbouring SSID's but out of curiosuity after seeing your message decided to run a few:
I use openspeedtest on the PC and connect to the rotuer with AX210 laptop

at 80Mhz im averaging (3 tests) around 740 down/500 up ping jumping between 80-40ms
at 160MHz 800 down, 600 up and ping 80ms

in the DFS channel at 160 wide I would normally get 970 down 940 up and forget the ping but withing normal expected range no what Ive quoted above

 

[sfx2000]

It's very unlikely that it's a switch issue, as Gigabit Ethernet switches haven't had any switching related problems for at least a decade by now. Also, the marketing on the switch you bought is across all ports and the switching IC inside of it, will only switch so much data per port.

I tend to agree - the switch element inside the SoC has more than enough bandwidth, even on the lower end SoC's like the Broadcom BCM6750 or the very similar Qualcomm IPQ4019...

I doubt it's the switch itself, and moving things out to a dedicated switch like hurts, not helps, as then the upstream for all traffic has to fall on to a single uplink port.

 

[sfx2000]

Curretnly I have 9 5GHz networks around me. 3 of them in the -40 to -50 range, two in the -60 range and all the rest in the -70 to -80 range. And all of them spanning across channels 36-52.

As long as you are 20 dBm higher than the neighbor, you're golden... just because you see a lot of AP's are a certain level, those levels are based on the Beacon Frames being transmitted at 6 Mbps (the lowest rate) and don't necessarily indicate what the channel utilization actually is.

FWIW 11n/ac/ax have co-channel interference level requirements that basically suggest that the neighbor AP could be up to +10 dBm stronger, and the AP/Client has to reject that noise....

 

[munkiemagik]

As long as you are 20 dBm higher than the neighbor, you're golden...

Thank you @sfx2000 , Ive been trying to find out information like this but havent come across any meniton of dbm level proximity with regards to interference. Theres always one but often its two and sometimes three SSIDs at all different times of the day that are well within 20 dbm range of me. Is that sufficient to warrant jumping onto the 6GHz bandwagon for VR streaming use case?

If I was just scrolling cat videos on youtube like th egood old days I woudlnt even be bothered wiht any of this. Its only the micrstutters in VR simraacing that got me fixated on network performance and hardware.

To address your prevous post, the AX58u does only have one SSID strictly reserved for the Quest 3 and no other wireless devices connect to the AX58u and 2.4GHz radio off.

 

[Tech9]

I've never run any Lan bandwidth tests with the AX58u on one of these occupied channels

You have to read about how Wi-Fi works first. What you see is radios tuned to the same frequency. They may not transmit any data at the moment and this same "occupied" channel may be in fact the one with the most available bandwidth. Usually most consumer routers are set on Auto and users don't know what Wi-Fi channel is. The channels you see "free" are the ones every router around you avoids for some reason. The available bandwidth is shared between all users. It changes dynamically, different at any moment. Your 5GHz radio dedicated to the VR set doesn't give you any advantages. If you need guaranteed high bandwidth in a dense Wi-Fi environment you have to move to 6GHz or find something else to play with. I also don't know anything about your Wi-Fi settings. Some may sound promising from advertisements, but in reality have negative effect on performance.