mirror of
https://github.com/Runnin4ik/dpi-detector.git
synced 2026-10-20 11:31:59 +00:00
ff90794a57
What the day's captures settled, written where the next pass can find it. The unopenable first reply is the local desync's fake, not an endpoint behaviour; the fake leaves with the flow's own 5-tuple and one DCID, so the edge holds a single half-open connection and its timeout is the gate. The gate is 2.5–3.5 s, not the 4.7 s the probe's ladder lands on: the probe's fourth send is the first after it, curl's third (3.5 s), and Chrome closes itself at 4.004 s of silence, half a second before its own next retransmit — so a browser cannot show the gate, only fail on it (`ERR_QUIC_PROTOCOL_ERROR`, the netlog is quoted in the section). The levers that would remove the gate are written down with what is measured and what is not (badsum, a TTL, a host list built by measurement, an unpadded fake), and the tester section points at the list the detector now ships. `replay.py` stops crashing on the long headers that are not Initials. |
||
|---|---|---|
| .. | ||
| quic | ||
| bench-strategies.sh | ||
| vendor-advisories.sh | ||