The quest for 500/500 over wifi

[Emixxary]

Hi all,
Would appreciate if any experts can take a look at my below experiments/reasoning and point out if I've missed obvious stuff or am just generally way off base.

Originally I had a Netgear R6400 router, and that was pretty much stable and easily maxed out my 30/30 mbit ISP plan. Then at some point I upgraded my ISP plan to 500/500, which promptly showed that my wifi connection wasn't capable of taking advantage of these speeds. Yes, I've read a lot and know not to expect miracles from wifi But I'm a bit of a geek and want to see how far I can push it (and also to understand the maze of confusing marketing numbers). Luckily, my set up has the router and PC about 20-30 ft apart with pretty clear line of sight other than a wooden railing between loft and living room. Less fortunately, there seem to be a fair number of other wifi networks nearby.

Setups I've tried so far:
1) Netgear R6400, Intel AC9260 client
2) Netgear R7800 (nighthawk x4s), Intel AC9260 client
3) Netgear R7800, Asus PCE-AC88 client

Setup 1: R6400 (4x4, unknown max channel width/QAM), Intel AC9260 (2x2, 160mhz/256 QAM)
With this setup I was getting ~165/260 mbit down/up at a signal strength of about -42dbm (as reported by InSSIDer). I played around with changing channels, but other than some channels being intermittently slower (I assumed due to interference) it didn't change much. After doing a lot of reading and consulting these wifi PHY speed charts https://docs.google.com/spreadsheets/d/1Ckfrx8lHCQuYKEiu6F-XUFAygobckSdFLTwGj1EpE_Y/edit#gid=0, I assumed I was getting one of the following:
2x2, 20mhz, 256-QAM 3/4
2x2, 40mhz, 16-QAM 3/4

I chalked this up to there being too many networks on the same channels and possibly signal strength being too weak; so I set out to get a "better" router that could use the DFS channels and hopefully take advantage of 80/160mhz channels. Enter setup 2!

Setup 2: R7800 (4x4, 160mhz/256 QAM), Intel AC9260 (2x2, 160mhz/256 QAM)
With the R7800 I got about the same signal strength, and was additionally able to use DFS channels. Disappointingly, switching to these channels didn't seem to have any effect on rates at all despite InSSIDer showing that there are zero other networks on the DFS channels, versus 6 ish "nearby" on the non-DFS ones. So I assume channel interference wasn't a major issue after all. Poking further into the settings I discovered a checkbox for "HT160" - 160mhz, score! I had pretty low hopes though, because I had assumed that already only 20/40mhz channels were being used... But this actually improved my rates to ~260 down, 400 up. This implied to me that 80mhz channels were actually being used with QPSK 3/4 and this bumped things up with double the channel width, but maybe dropping MCS to QPSK 1/2.

Armed with this knowledge, I figured - if I'm getting 260mbit over a 2x2 connection limited by my motherboard's onboard wifi and sharkfin antenna, I should be able to easily get 520mbit with a fancier adapter using a 4x4 connection and (presumably) better signal strength for equal or better MCS. Enter setup 3!

Setup 3: R7800 (4x4, up to 160mhz/256QAM), Asus PCE-AC88 (4x4, 80mhz/1024 QAM)
Well, I should have read the specs better - it turns out the PCE-AC88 doesn't support either DFS channels or 160mhz (which I guess makes sense since it looks like there aren't 160mhz-wide contiguous ranges of non-DFS channels). I am now getting a -32dbm signal according to InSSIDer, roughly a 10dbm improvement; but the end result is a 'disappointing' 330mbit down/480 up. Mathematically it makes sense - it exactly doubles the 2x2 80mhz rates. What I'm a bit puzzled about is why the MCS is so low.

Based on a chart graphic here: https://7signal.com/802-11ac-migration-part-1-what-nobodys-telling-you-about-256-qam/

It appears to me that with -32dbm signal strength I should be well within the range to get 5/6 256 QAM on 160MHz (45-50dbm on that chart), which is 1733mbit over 2x2. Or alternately, 5/6 256 QAM on 80MHz over 4x4, which should also be 1733mbit. Yet I'm getting rates that actually look like 1/2 or 3/4 QPSK, which is 7-8 MCS indexes lower. This holds true on both the 'crowded' normal channels, and the completely unused DFS channels.


So ultimately I have the following questions:
- What am I missing about MCS/rates vs. signal strength/interference?
- Are there any other avenues to pursue for increasing throughput, other than going wired or moving closer to the router?


Thanks!

 

[psychopomp1]

- Are there any other avenues to pursue for increasing throughput, other than going wired or moving closer to the router?

Yes. Get a wifi 6 router and pair that up with a wifi 6 PCIe card.

 

[Emixxary]

Yes. Get a wifi 6 router and pair that up with a wifi 6 PCIe card.

Well, should have seen that response coming.

My understanding from what I've read though is that wifi6's main benefits come from the following:
- better mu-mimo and ofdma - only benefits when multiple devices are consuming large amounts of bandwidth simultaneously
- standardized 160mhz
- higher max MCS

Now maybe this ties back in to what I'm not understanding - am I really only getting mcs 2-3 on my current setup? If I am, is there any reason to believe I'd get any benefit out of the higher mcs tiers in wifi 6?

If not, then presumably I'd be stuck back with the same rates as I was getting on 2x2 160mhz, a downgrade from my current 4x4 80mhz rates.

 

[thiggins]

AX uses the same maximum 1024 QAM that Broadcom has supported in its AC chipsets, but it also supports it in 2.4 GHz and the top rate of MCS 11 is now a part of 802.11.

Your best bet to achieve the rate you want is to use another R7800 in bridge mode and 160 MHz channels. -32 dBm is a pretty damn good signal level for 5 GHz, but keep in mind RSSI readings are not necessarily "truth" since they usually are not calibrated.

AX routers might provide a slight edge due to newer radio designs, but you'd need a pair of 4 stream routers, which is an expensive move.

Aside from being able to make full use of the ISP bandwidth you are paying for, do you really need more than 300 Mbps throughput?

 

[Emixxary]

Aside from being able to make full use of the ISP bandwidth you are paying for, do you really need more than 300 Mbps throughput?

Absolutely not This is half OCD chasing of theoretical limits/pretty numbers, and half just trying to figure out how things work and penetrate the veil of marketing hype/obfuscation so I can understand what the the real limits are given how much money/effort.

**a couple hours later**
OKAYYYYYY so I had a bunch of follow up questions typed up about SNR, bridge/mesh setups and expectations for backhaul etc... but I accidentally stumbled into the realization that I'm a complete idiot. The short story: I should have paid more heed to warnings that WAN based speed testing is fraught with issues.

Longer story:

but keep in mind RSSI readings are not necessarily "truth" since

Based on this, I started geeking out about RSSI and noise, and stumbled across the following tidbits:
a) MCS is determined by SNR, not RSSI - and SNR is RSSI minus noise, so to get an accurate picture I'd need to find out what my noise levels are.
b) A video talking about how getting a bigger antenna can actually hurt SNR by picking up more noise sources, and you're better off getting a directional antenna with lower gain that can pick out just the targeted router.
This tickled my brain into remembering that I saw a 'beamforming' option in my Asus client card's device settings, which had been disabled by default. I enabled it with a healthy dose of skepticism, the card rebooted itself and reconnected to the network, and... BAM! instant 600mbit peak, tailing off to 480mbit. For science, I disabled the setting and re-ran the speed test (yes, speedtest.net like the idiot I am). To my surprise I got the same identical result. I noticed that the server was now 'accretive networks' instead of 'ziply fiber', and re-ran the test against Ziply. This promptly gave me my old result of 300 ish mbit.

So I guess the moral of the story is: Don't assume speedtest.net servers can actually keep up with your internet plan's bandwidth, *even if the server is your own ISP*. And now I'm extremely annoyed because I thought I'd been close to figuring it all out, and now I need to go back and do a lot more experimentation because apparently my test method was completely broken. And yes, my 2x2 160mhz setup also apparently hits the 500mbit mark against this server, so now I'm scratching my head trying to figure out if my prior results were from hitting different servers purely by chance, despite doing repeated trials. Sigh. Luckily I'd also ordered a gigabit ethernet adapter for my laptop because I wanted to test hard-wired rates (I actually never ended up using it, because I discovered that my router can do a speed test itself and that was showing 500/500 to the ISP). So I guess I can set up a proper test on my LAN now and repeat my experiments.

 

[thiggins]

Some speedtest sites are better than others. Anything works with my measly 10/0.8 DSL connection. But as you get into the 100's of Mbps connections, you need to be choosier.

Best option is to use iperf3 with one end being a LAN machine connected via gigabit Ethernet. I should have suggested this in my first response.

 

[Emixxary]

Best option is to use iperf3 with one end being a LAN machine connected via gigabit Ethernet

Thanks, I'll give this a shot. I was just about to google "how to benchmark LAN speeds"

 

[degrub]

Yeah, iperf for anything on your side of the isp modem.

 

[Tech Focus]

You don't need 160MHz channel width nor Wi-Fi 6 to get up to 500 Mbps single device performance. The key element is Wi-Fi system efficiency and PHY. Once you have 2x2 MIMO client @ 80 MHz channel width, your R7800 should be able to make 866 MHz PHY. Not only signal strength but noise is important because Signal to Noise Ratio (SNR) determines the link speed so if you have -30 dBm but noise is -50, you have only 20 SNR and that won't get you the peak speed.

If the system is 60% efficient, you will hit 500 Mbps. Otherwise, 3x3 client e.g. some laptops (my case is MacBook Pro), I can get 700-800 Mbps online speed test using my Wi-Fi 5 Wave 2 AP system.

As you've asked distance plays also good factor, but having said this if you have no obstacle, certain distance are much more forgiving. But remember obstacle includes not only walls but human, sofa, furniture depending on where you place your router. Hence, high placement of router is often recommended. In fact 20-30 ft, I did my own test in my home, I still got over 500 Mbps online speed test result with my system.

The bit tricky part of your data is asymmetric down & up with down being lower than up. With my limited experience of reading across internet, this issue is often tricky to find out the core of issue. I solved mine by going step by step in my system and found our my case is SFP+ port/link between switch and router, but this seems possibly way wider range of problem lists. The most obvious is you want to make sure channel utilization is not too high, try couple different speed test online sites (or as you are trying first solving iPerf 3), and ensure no competing traffic on downlink side.

One last piece, is be sure to check speed at router. For instance for 1 Gbps speed, Xfinity says 940 Mbps down is what they are promising (at good times). Depends on the time of day, I get even under 800 Mbps at router. So if same thing is happening to you, your ISP may not be giving you constant 500 Mbps.

 

[Emixxary]

Hmm, was getting some weird results with iperf - as in, 160mbit via iperf (LAN) versus 450mbit via speedtest.net (WAN). Maybe it has something to do with not using parallel streams? But it doesn't seem to connect if I specify to use parallel streams.

I ended up just creating a 1gb file and transfering it via File Explorer... ghetto, but seemingly effective. According to File Explorer, I'm now seeing the following:
- 80MByte (640MBit) via 2x2 80mhz
- 108MByte (864 MBit) via 2x2 160mhz
- 108MByte (864 MBit) via 4x4 80mhz

Given that the gigabit adapter on the other end is the cheapest USB dongle I could find on Amazon, I'm guessing that's the bottleneck for the latter results rather than the wifi speeds, as I'd probably expect going from 2x2 to 4x4 to be more than a 30% speed increase. Speedtest.net against the 'fast' server corroborates these results with close to 500mbit over 2x2 80mhz. So ultimately I guess I really didn't need to spend any of this money

That said, I'll be curious to repeat these tests throughout the week and see if the rates hold or if they drop when people come home from work and start streaming 4k netflix on adjacent networks or some such. Because I swear to god there was a pretty clear correlation between the different setups and WAN speeds :/

 

[thiggins]

Make sure you are using iperf3. Download the latest version from here. Man page is here.

Start the iperf3 server on the machine connected via Ethernet to a router LAN port. (iperf3 -s)

Start the iperf3 client on your wireless machine. Using the TCP/IP defaults is fine (iperf3 -c [ip address of the iperf3 server]). Traffic runs from client to server, so that would be uplink. For downlink, add the -R switch. To run longer add -t.

So to run a 60 second TCP/IP downlink test with a server @ 192.168.1.1 with 5 parallel connections, the command on the client machine would be:
iperf3 -c 192.168.1.1 -t 60 -R -P 5

 

[Emixxary]

Ah, I got it from iperf.fr which looks like it's a couple years out of date. I'll give that a shot later, thanks.

 

[doczenith1]

So I guess the moral of the story is: Don't assume speedtest.net servers can actually keep up with your internet plan's bandwidth, *even if the server is your own ISP*.

I'll add to this in that even with the same server I get wildly different results. Computer is plenty capable i5-9600K with SSD running Windows 10.

Here is the result from their website test

And the exact same server using the CLI in Windows command promt.

Using iPerf3 to test wireless throughput I get the best results adding the -w 4M option. My input string is: iperf3 -c 192.168.1.11 -O 5 -i 5 -w 4M -t 60

 

[Makaveli]

I'll add to this in that even with the same server I get wildly different results. Computer is plenty capable i5-9600K with SSD running Windows 10.

Here is the result from their website test
View attachment 24430

And the exact same server using the CLI in Windows command promt.
View attachment 24429


Using iPerf3 to test wireless throughput I get the best results adding the -w 4M option. My input string is: iperf3 -c 192.168.1.11 -O 5 -i 5 -w 4M -t 60

I had no idea there was a Cli version of speedtest available until I read your post thanks.