KIARA / MEASUREMENTS
The cube under traffic.
Two hours of traffic through the C126 test cube in four modes: without the cube, filter only, VPN, and VPN with the filter. We show the numbers together with what this measurement does not tell you.
- of warning-list domains blocked1,476 of 1,476 lookups of test domains
- 100%
- download with the filter onlywithout the cube 9.8 Mb/s: the filter barely slows things down
- 9.2 Mb/s
- download through the VPN with the filterfull protection at once; without the cube 9.8 Mb/s
- 6.3 Mb/s
- VPN connections went through the tunnelnone went around the tunnel
- 324 of 324
Four modes
| Mode | Download, Mb/s | Upload, Mb/s | Latency | DNS: answers · new name | Warning-list domains | Restarts |
|---|---|---|---|---|---|---|
| Without the cube | 9.8 Mb/s | 10.5 Mb/s | 47 ms | 162/172 · 29 ms | 495/495 answered | 0 |
| Cube: filter only | 9.2 Mb/s | 8.0 Mb/s | 49 ms | 168/168 · 34 ms | 504/504 blocked | 0 |
| Cube: VPN ad filter paused | 6.3 Mb/s | 3.3 Mb/s | 50 ms | 162/162 · 69 ms | 486/486 blocked | 0 |
| Cube: VPN + filter | 6.3 Mb/s | 3.4 Mb/s | 50 ms | 162/162 · 69 ms | 486/486 blocked | 0 |
Median of the cycles in each mode. The tick on each bar is our minimum threshold, set before the phone tests: 5 Mb/s down and 1 Mb/s up. A “new name” is a lookup of a random domain, which goes through the whole chain of DNS servers.
Speed over time
Download above the axis, upload below it. Each strip covers one mode and uses the same 0–12 Mb/s scale, so the strips can be compared at a glance. The capped lines are our minimum thresholds.
Without the cube
median: download 9.8 · upload 10.5 Mb/sCube: filter only
median: download 9.2 · upload 8.0 Mb/sCube: VPN
median: download 6.3 · upload 3.3 Mb/sCube: VPN + filter
median: download 6.3 · upload 3.4 Mb/sDNS: an answer or silence
Each cell is one cycle: three lookups of ordinary, allowed domains. A full cell means all of them answered, a sparser one that some did, and an empty frame that none did. Dots at the end of a row are cycles this mode did not have. A copper cell is a cycle in which the lookup of a new name took longer than 300 ms. This measurement has none.
The cube's memory
Free internal memory of the cube through the whole run, read every cycle. The copper line is the lowest level since start-up: it steps down when the cube starts carrying traffic. The numbers on the chart match the modes below it.
- Least free memory
- 27.6 KB
Memory does not leak away: the lowest level settled at about minute 83 and did not drop further until the end of the measurement.
- 1Without the cube
- 2Cube: filter only
- 3Cube: VPN
- 4Cube: VPN + filter
Does VPN traffic leave through the tunnel
The server records where each measurement connection came from: the cube's VPN tunnel or the home connection. In the VPN modes every connection should arrive through the tunnel.
| Mode | Through the tunnel | Home connection | Another way | Result |
|---|---|---|---|---|
| Without the cube | 0 | 162 | 0 | plain connection, as expected |
| Cube: filter only | 0 | 168 | 0 | plain connection, as expected |
| Cube: VPN | 162 | 0 | 0 | all through the tunnel |
| Cube: VPN + filter | 162 | 0 | 0 | all through the tunnel |
How we measured
- Each cycle is 15 seconds of download, 15 seconds of upload, 20 latency samples and DNS lookups: a new random name, an ordinary domain and three test domains from the warning list.
- The client is an M5Stack Core2 board running our own measurement program. Without the cube it joins the home Wi-Fi; in the other modes it joins the cube's network.
- The cube and the client are connected to a computer by cable only to collect the numbers. The measurement server is our own test server, not a VPN provider's server.
What this measurement does not tell you
- The ceiling of this measurement is the client itself: a Core2 board with a 64 KB TCP window, talking to a server about 50 ms away, cannot go much past 10 Mb/s, although the home line has 2 Gb/s and the router is 4 metres away. Without the VPN the cube stayed under that ceiling, so in this measurement it was not the bottleneck.
- With the VPN the cube gives about 6 Mb/s down and 3 Mb/s up, below the client's ceiling, so that is its current limit in this mode. Upload is the weakest part. We are looking into the cause: the candidates are WireGuard encryption computed on the cube's processor and its Wi-Fi buffers.
- The VPN tunnel went to our test server. Through a VPN provider's server, Proton for example, latency and speed also depend on its network.
- In the “VPN” mode the ad filter was paused, but the CERT warning list always applies, so the test domains are blocked there too.
- We have not yet measured faster clients such as a phone or a laptop. One cube, one home Wi-Fi, one server: this is a lab measurement, not a speed promise for everyone.
Runs
- perf-20261017T154430Z: modes Without the cube, Cube: filter only, Cube: VPN, Cube: VPN + filter, cube image 06429f1f0ff4…, start 2026-10-17T15:44:40Z