[Make-wifi-fast] Real router latency comparison

[Make-wifi-fast] Real router latency comparison

[Tim Higgins]

[-- Attachment #1: Type: text/html, Size: 1873 bytes --]

[-- Attachment #2: four_router_latency_compare_up.png --]
[-- Type: image/png, Size: 156919 bytes --]

[-- Attachment #3: four_router_latency_compare_dn.png --]
[-- Type: image/png, Size: 146406 bytes --]

[-- Attachment #4: four_router_congestion_compare_dn.png --]
[-- Type: image/png, Size: 176813 bytes --]

[-- Attachment #5: four_router_congestion_compare_up.png --]
[-- Type: image/png, Size: 150420 bytes --]

Re: [Make-wifi-fast] Real router latency comparison

[Dave Taht]

Tim:

well, can you run the test with the _be test, not mixed dscps, also?

Also are you getting actual latency differences from using each dscp?
or summing it?

A lot of vendors don't differentiate at all. individual stream plots
out of this data would show this, no difference in latency, across
mixed dscps. in that case, one vendor actually following the 802.11e
spec, and others not, hmmm.

But: I've always recommended openwrt turn off 802.11e differentiation
on the AP, and this is why. It is VASTLY better to optimumize for
802.11n aggregation than 802.11e. on the AP. Clients, sure, try it....

I know/knew exactly what the two things wrong here, but never got
around to fixing it, as in practice dscps are hardly used at all.

1) we use up a txop in all the queues rather than just 1 at a time.
2) the codel algorithm has some weird code in it the mucks with the
target when it shouldn't and we use a too large target for 5ghz.

Still not bad for a 2x ap against bigger fatter tech that I personally
have never seen.

THX FOR THE TEST!!!


On Sun, May 17, 2020 at 11:47 AM Tim Higgins <tim@smallnetbuilder.com> wrote:
>
> Hi all,
>
> Attached are previews of results I'll be publishing in the Part 2 article.
>
> The same test was run on four routers:
> - NETGEAR R7800 (Qualcomm 4 stream AC)
> - Evenroute IQrouter v3 (Mediatek 2 stream AC, Wifi stack is "vanilla"OpenWRT for the MT76" according to Evenroute)
> - NETGEAR RAX120 (Qualcomm four stream AX) - OFDMA enabled
> - NETGEAR RAX45 (Broadcom four stream AX) - OFDMA enabled
>
> AP set to Channel 36, 80 MHz bandwidth, WPA2 PSK connection
>
> Each STA shows 867 Mbps link rate. Signal level ~ -45 dBm.
>
> iperf3 TCP/IP traffic run simultaneously to all four STAs (Intel AX200, Win 10, 21.80.2.1 driver)
> bitrate  (-b): 50Mbps
> length (-l): 256 Bytes
> DSCP values (--dscp), one per STA: 0 (CS0), 96 (CS3), 160 (CS5), 192 (CS6)
>
> 200ms interval ping run concurrently from AP to STA on each pair. Ping is always AP to STA for both uplink and downlink traffic.
>
> Congestion was measured using a Qualcomm AX STA associated to the AP, running 1bps of traffic so that stats could be recorded.
>
> Thoughts?
>
> ===========
> Tim
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast



--
"For a successful technology, reality must take precedence over public
relations, for Mother Nature cannot be fooled" - Richard Feynman

dave@taht.net <Dave Täht> CTO, TekLibre, LLC Tel: 1-831-435-0729

Re: [Make-wifi-fast] Real router latency comparison

[Dave Taht]

On Sun, May 17, 2020 at 11:57 AM Dave Taht <dave.taht@gmail.com> wrote:
>
> Tim:
>
> well, can you run the test with the _be test, not mixed dscps, also?
>
> Also are you getting actual latency differences from using each dscp?
> or summing it?

An example of where you can go wrong by summing things was in my
old make-wifi-fast preso here:

where given the 3-10 sec of latency built into the ath10k driver at the time
5 flows grabbed all the bandwidth, and the other 95 failed to start entirely.

https://blog.linuxplumbersconf.org/2016/ocw/system/presentations/3963/original/linuxplumbers_wifi_latency-3Nov.pdf

>
> A lot of vendors don't differentiate at all. individual stream plots
> out of this data would show this, no difference in latency, across
> mixed dscps. in that case, one vendor actually following the 802.11e
> spec, and others not, hmmm.

On the other hand, it is certainly possible that ofdma and DU have
finally got to where they could priority schedule
stuff. you'd have to look at the air....

>
> But: I've always recommended openwrt turn off 802.11e differentiation
> on the AP, and this is why. It is VASTLY better to optimumize for
> 802.11n aggregation than 802.11e. on the AP. Clients, sure, try it....
>
> I know/knew exactly what the two things wrong here, but never got
> around to fixing it, as in practice dscps are hardly used at all.

Three things, sorry.

0) Using anything other from the BE queue.

This can be fixed with a proper qos-map value in the ap's conf file. I
don't know how to do that
in openwrt yet - I always just patched it out.

> 1) we use up a txop in all the queues rather than just 1 at a time.
> 2) the codel algorithm has some weird code in it the mucks with the
> target when it shouldn't and we use a too large target for 5ghz.
>
> Still not bad for a 2x ap against bigger fatter tech that I personally
> have never seen.
>
> THX FOR THE TEST!!!

And ping is not the best test. Synthetic voice or videoconferncing
traffic would be better. irtt might be
something worth using or some other rtp or webrtc test suite.

Still, I expected to lose on this, but we didn't lose badly, and
unlike those other vendors,
evenroute and openwrt folk *listen* and do what they can to improve things.

>
> On Sun, May 17, 2020 at 11:47 AM Tim Higgins <tim@smallnetbuilder.com> wrote:
> >
> > Hi all,
> >
> > Attached are previews of results I'll be publishing in the Part 2 article.
> >
> > The same test was run on four routers:
> > - NETGEAR R7800 (Qualcomm 4 stream AC)
> > - Evenroute IQrouter v3 (Mediatek 2 stream AC, Wifi stack is "vanilla"OpenWRT for the MT76" according to Evenroute)
> > - NETGEAR RAX120 (Qualcomm four stream AX) - OFDMA enabled
> > - NETGEAR RAX45 (Broadcom four stream AX) - OFDMA enabled
> >
> > AP set to Channel 36, 80 MHz bandwidth, WPA2 PSK connection
> >
> > Each STA shows 867 Mbps link rate. Signal level ~ -45 dBm.
> >
> > iperf3 TCP/IP traffic run simultaneously to all four STAs (Intel AX200, Win 10, 21.80.2.1 driver)
> > bitrate  (-b): 50Mbps
> > length (-l): 256 Bytes
> > DSCP values (--dscp), one per STA: 0 (CS0), 96 (CS3), 160 (CS5), 192 (CS6)
> >
> > 200ms interval ping run concurrently from AP to STA on each pair. Ping is always AP to STA for both uplink and downlink traffic.
> >
> > Congestion was measured using a Qualcomm AX STA associated to the AP, running 1bps of traffic so that stats could be recorded.
> >
> > Thoughts?
> >
> > ===========
> > Tim
> > _______________________________________________
> > Make-wifi-fast mailing list
> > Make-wifi-fast@lists.bufferbloat.net
> > https://lists.bufferbloat.net/listinfo/make-wifi-fast
>
>
>
> --
> "For a successful technology, reality must take precedence over public
> relations, for Mother Nature cannot be fooled" - Richard Feynman
>
> dave@taht.net <Dave Täht> CTO, TekLibre, LLC Tel: 1-831-435-0729



-- 
"For a successful technology, reality must take precedence over public
relations, for Mother Nature cannot be fooled" - Richard Feynman

dave@taht.net <Dave Täht> CTO, TekLibre, LLC Tel: 1-831-435-0729

Re: [Make-wifi-fast] Real router latency comparison

[Tim Higgins]

[-- Attachment #1: Type: text/html, Size: 1363 bytes --]

[-- Attachment #2: iqrouter_vs_r7800_cdf_per_pair_dn.png --]
[-- Type: image/png, Size: 192851 bytes --]

[-- Attachment #3: iqrouter_vs_r7800_cdf_per_pair_up.png --]
[-- Type: image/png, Size: 168276 bytes --]

[-- Attachment #4: iqrouter_vs_rax120_cdf_per_pair_dn.png --]
[-- Type: image/png, Size: 189503 bytes --]

[-- Attachment #5: iqrouter_vs_rax120_cdf_per_pair_up.png --]
[-- Type: image/png, Size: 194178 bytes --]

Re: [Make-wifi-fast] Real router latency comparison

[Tim Higgins]

[-- Attachment #1: Type: text/html, Size: 1306 bytes --]

[-- Attachment #2: rax120_mixed_vs_all192_dscp_dn.png --]
[-- Type: image/png, Size: 159900 bytes --]

[-- Attachment #3: rax120_mixed_vs_all192_dscp_up.png --]
[-- Type: image/png, Size: 177884 bytes --]