Cutie makes a Quality Test of your network. It continually tests Network Quality (MOS - a measure of the quality of a voice or video call), Packet Loss, and Latency/Jitter.
Any impairments to the network appear in the graphs. Cutie hardly uses any resources - a little less than two kilobytes per second. Here's a screenshot with description below:
The charts above were created using a cell phone hotspot: its relatively low speed link magnifies any impairments. Here are the highlights.
- The first few minutes (up to 10:37) show no-load conditions: no packet loss and stable latency of about 90msec. This leads to excellent Network Quality.
- Around 10:39, begin a lively video on another computer. Note the brief packet loss (likely caused when starting the Youtube video) leads to dips in Network Quality. The latency and jitter remained fairly stable.
- Around 10:42, run a speed test on that other computer. Notice that latency jumps up very high (over 250msec) for the duration of the speed test with a couple brief instances of packet loss. These cause the MOS Quality to plummet during the test.
Cutie's WebRTC connection creates fine-grained measurements by sending 10 probes per second, (every 100 ms). The three charts display the 10-second average of the values below:
-
Network Quality (MOS). The Mean Opinion Score is an industry standard that expresses the quality of a voice call (and by extension, of a videoconference call). The MOS calculation produces values between 4.5 (excellent) and 1.0 (bad) and is computed using the packet loss, latency, and jitter measurements.
-
Packet Loss (%) Cutie determines packet loss by detecting missing sequence numbers from the stream of echoed messages.
-
Latency & Jitter Cutie computes the difference between time the message was received and the timestamp within the message to determine the latency for each message. It computes the jitter from the differences between arrival times of subsequent messages.
Note that Cutie measures the performance of the entire network. Traffic from other computers can affect the measurements in the charts.
You can try Cutie from the demo site at https://cutie.richb-hanover.com
Notes:
- The base ICMP ping time to that server is about 35 ms (it's a pretty slow VPS). Consequently, Cutie's lowest latency tends to be about 35 ms.
- You could also install a Cutie server on a nearby computer to test your local network's abilities.
It is straightforward to install this code on Linux or macOS:
git clone https://github.com/richb-hanover/Cutie-network-Quality-Test.git
cd Cutie-network-Quality-Test
npm installTo run the code:
# to test from localhost:5173
npm run dev
# To leave a test server running
nohup npm run dev &To start a production server, read the CHECKLIST for details.
As noted in the Theory of Operation,
this was initially developed in VSCode with the Codex LLM plugin.
Subsequent versions were developed using obra/superpowers
in the Warp terminal emulator.
- I envision bundling
coturninto a Docker container at some point to create a turnkey server along with netperf, iperf, iperf3, Crusader server, etc. This container could run on something small (like a Raspberry Pi 4) that could be dropped onto a network anywhere and used as a test platform. - There are occasional surprises (errors) in the reported numbers as compared to values recorded in the charts.
It's a vaguely humorous pronunciation (for English speakers) of "QT" for "Quality Test". A saving grace is that "Cutie" seems not to have collisions in a quick Google search for "cutie network test".
NB: I am aware of the QT Graphics View Framework that shares the same acronym "QT". However, they pronounce it "cute".
This is version 0.3.1, and is beta-test quality code.
Read the
Provenance - Vibe Engineering
information to see how this was derived,
as well as the obra/superpowers documents in docs/plans.
I would be pleased to get feedback or bug reports on the Issues page.