[Bloat] Fwd: Ookla Speedtest app on iPad with Thunderbolt Ethernet

[Dave Taht]

---------- Forwarded message ---------
From: Stuart Cheshire <cheshire at apple.com>
Date: Tue, May 31, 2022 at 10:33 AM
Subject: Ookla Speedtest app on iPad with Thunderbolt Ethernet
To: Dave Täht <dave.taht at gmail.com>, Sasha Leitmann
<sashaleitmann at gmail.com>, Matt Mathis <mattmathis at google.com>,
Jonathan Foulkes <jfoulkes at evenroute.com>, Christoph Paasch
<cpaasch at apple.com>


I got hold of an iPad with Thunderbolt to try the new Ookla Speedtest
app with Gigabit Ethernet. Here are my results.

                    Download    Upload   Idle   Idle   Download
Download Upload  Upload
Network adapter SQM Throughput Throughput Ping Ping Max   Ping   Ping
Max  Ping  Ping Max

Apple    USB 2  No      95        35.3     17     21        72
459      58    1668
Moshi    USB 3  No     917        39.3     16     19        84
281      22    1007
Mokin    USB C  No     910        36.2     16     19       105
374      62     764
StarTech USB C  No     890        39.0     16     20       154
744      25     852

Apple    USB 2  Yes     85        20.9     17     19        10
39      12      22
Moshi    USB 3  Yes    689        20.9     15     20        10
46      12      21
Mokin    USB C  Yes    723        21.1     16     19        10
30      10      19
StarTech USB C  Yes    705        20.2     17     19        10
34      11      18

None (Wi-Fi)    No     425        37.4     18     30       159
638      63    1276
None (Wi-Fi)    Yes    204        21.2     18     24        22
49      14     127

The columns to pay attention to are the Download and Upload Ping times.

For the Ethernet connections, when turning on SQM:

Download Ping     goes from around 100 ms to around 10 ms, a 10x improvement.
Download Ping Max goes from around 400 ms to around 40 ms, a 10x improvement.

Upload Ping     goes from around   40 ms to around 12 ms, a  3x improvement.
Upload Ping Max goes from around 1000 ms to around 20 ms, a 50x improvement.

The figures for Wi-Fi are, unsurprisingly, a little worse overall than
wired Ethernet, but still show 10x improvement when SQM is used.

The reduction in delay, particularly in the upload direction, is quite dramatic.

Also bear in mind that I’m using Comcast cable modem service here, and
Comcast has already deployed better delay management than most ISPs.
For details, see “Improving Latency with Active Queue Management (AQM)
During COVID-19” <https://arxiv.org/abs/2107.13968>

People with less advanced ISPs would likely see even bigger gains from
applying competent SQM.

Stuart Cheshire

P.S. I set my artificial throughput limits conservatively low to be
sure I was seeing the SQM queue management effects, so there’s a
noticeable reduction in throughput. In a real deployment where the
Cable Modem and CMTS are doing smart queue management themselves you
wouldn’t see this reduction in throughput because the cable hardware
has direct knowledge of the instantaneous rates and queue sizes. When
I’m doing this artificially with the UDM Pro, I have to work with
pessimistic worst-case throughput limits to be certain I won’t
accidentally end up with a queue building up elsewhere.


-- 
FQ World Domination pending: https://blog.cerowrt.org/post/state_of_fq_codel/
Dave Täht CEO, TekLibre, LLC