Transport Layer Manipulation
Format. Break into groups of 4–5. Each group picks one of the two breakouts below (or takes both if time allows). Spend ~5 minutes skimming the prep reads, then ~10 minutes debating. A designated reporter brings the group’s position — and any dissents — back to the full class for a ~3-minute report-back.
The transport-layer story in §2.2.2 is a fight over metadata. TCP RSTs, SNI-based filtering, and DPI on the Great Firewall all work because the envelope is legible even when the contents are encrypted. Encrypted ClientHello (ECH), QUIC, and TLS 1.3 change that. These breakouts ask whether protocol designers should be picking sides — and whether the censor’s side has already lost.
Breakout A: Should Protocol Designers Pick Sides?
Motion. “IETF working groups on ECH, QUIC, and DNS-over-HTTPS are effectively anti-censorship activism dressed up as engineering, and that is a good thing.”
Prep reads (5–10 min).
- Russia’s internet watchdog blocks thousands of websites that use Cloudflare’s privacy service — The Record, November 2024. Roskomnadzor blocks any TLS handshake carrying
cloudflare-ech.com as its outer SNI, one month after Cloudflare’s ECH re-launch.
- Encrypted Client Hello (ECH) in Censorship Circumvention — Niere et al., FOCI 2025. Measurement paper showing ECH deployment is essentially Cloudflare-only, and censors can defeat it by blocking the small set of advertised outer SNIs.
- Do Not Stick Out: The Dynamics of the ECH Rollout — Center for Democracy & Technology, 2025. Frames the paradox: ECH only works when it is ubiquitous, but to get ubiquitous it has to start somewhere — and early adopters get picked off.
- Disrupted, Throttled, and Blocked: State Censorship in Russia — Human Rights Watch, July 2025. Documents Cloudflare’s, Mozilla’s, Apple’s, and Google’s responses to Russian pressure over ECH, VPNs, and app-store takedown demands.
Discussion prompts.
- The book’s Turkey and Great Firewall examples both depend on the censor seeing something in the clear — a DNS query, an SNI, an IP. TLS 1.3 with ECH closes the SNI hole. Is protocol standardization the right venue to make that choice, or is this a policy decision being smuggled into a technical body?
- Enterprise network admins and school-district filters use the exact same visibility that censors do. Chapter 1 flagged that the tools are the same; only the operator differs. Should ECH ship with a documented enterprise-bypass mode, or is that a backdoor by another name?
- QUIC runs on UDP and is much harder to reset with a mid-stream injection than TCP. If you were designing GFW v2, what would you do about QUIC? Block it entirely? Force downgrade to TCP? Something else?
- If protocol design becomes an explicit venue for anti-censorship advocacy, what stops the same body from being captured by governments that want the opposite (see: earlier IETF fights over lawful intercept)? Is a “neutral” IETF even coherent?
Bring back. One design decision (in ECH, QUIC, or DoH) your group would reverse, and why.
Breakout B: Is the Cat-and-Mouse Game Winnable?
Motion. “Deep packet inspection has already lost: within five years, no state-level censor will reliably distinguish circumvention traffic from ordinary HTTPS.”
Prep reads (5–10 min).
- Exposing and Circumventing SNI-based QUIC Censorship of the Great Firewall of China — GFW Report, USENIX Security 2025. Since April 2024, the GFW decrypts QUIC Initial packets at line rate to recover the SNI — but SNI fragmentation across CRYPTO frames sidesteps it.
- Staying ahead of censors in 2025: What we’ve learned from fighting censorship in Iran and Russia — Tor Project, 2025. Snowflake and WebTunnel are absorbing the load as obfs4 gets fingerprinted; Conjure ships to defeat bridge-list enumeration.
- Snowflake-targeted DTLS filtering in Russia, starting 2026-03-30 — net4people/bbs, March 2026. Russia begins DTLS ja3/ja4 fingerprint blocking of Snowflake after months of near-100% success, marking a fresh escalation.
- Iran’s January 2026 Internet Shutdown: Public Data, Censorship Methods, and Circumvention Techniques — arXiv, 2026. Layered DNS + HTTP(S) + protocol-level filtering during the January 2026 blackout, plus what actually kept working (obfs4, Snowflake, TLS fragmentation).
Discussion prompts.
- The book calls this “a continual cat-and-mouse game.” What would it take for that game to actually end? Does either side have a plausible endgame, or is this a permanent standoff?
- GFW-style censors have started using machine-learning classifiers on traffic patterns (packet sizes, timing) rather than payload contents. If content is encrypted but shape is legible, has encryption really helped? What would “traffic-shape indistinguishability” cost in performance?
- The book notes that some censors don’t beat obfuscation technically — they beat it by coercing VPN providers into compliance or by criminalizing their use. If the technical game is a stalemate, does the fight move entirely into the legal/economic layer?
- Rank these circumvention tools by five-year survivability against a well-resourced censor: commercial VPNs, Tor with obfs4, Shadowsocks, Snowflake, domain fronting via CDNs. Defend the top and the bottom.
Bring back. Your group’s prediction on the state of censorship-circumvention in China (or another named censor) in 2030, in one sentence.
Instructor notes
These map to the “Defending against TCP/IP manipulation” subsection and the section takeaway that encryption is “pushing censors toward coarser tools.” Breakout A works better with students who have IETF/W3C exposure or who have thought about standards politics. Breakout B is more concrete and better for students who want to argue about specific tools. Consider running B first as a warmup and A as the more philosophical follow-on — the ECH/QUIC debate has more edges once students have named the tradeoffs in the arms race.