Best VPN Protocol & Deployment for Russia — Final Report (October 2026)

Status: research synthesis as of 2026-10-25. Russia's censorship (ТСПУ / TSPU) changes week to week. Treat every recommendation as a starting point and test on your own ISP / mobile operator. Method: six rounds of research across Xray-core / net4people GitHub threads, Habr, te-st.org, zapret wiki, GFW Report (academic), and Iranian / Chinese circumvention experience. Confidence levels are given for each claim. 简体中文版: https://gist.github.com/nothing-holding/d7ac4d7f5ff9fd4132730c0577c0324b

TL;DR

Question Answer
Best protocol VLESS + XHTTP (stream-one) + REALITY, with VLESS Encryption + XTLS Vision, XMUX limited to 2 connections, uTLS fingerprint firefox, Xray-core ≥ v26.10.21
Best speed profile VLESS + TCP + REALITY + Vision (firefox), home broadband only
Best backup Hysteria2 (Salamander + port hopping, BBR) on a different country / ASN
Best hosts AlphaVPS Nuremberg (€2.99/mo, primary) + UpCloud Helsinki (€3/mo, backup) — both passed live probes from 28 Russian vantage points, clean ASNs (see §4)
Mobile "whitelist" mode Russian whitelisted-subnet relay → foreign exit (measured 26–94 Mbit/s); emergency: WebRTC video-call tunnel, DNS tunnel
Dead / avoid Shadowsocks (incl. Outline), WireGuard, OpenVPN, IKEv2/L2TP, SOCKS5, MTProto, ShadowTLS v3, default chrome fingerprint, Hetzner / DigitalOcean / OVH / Vultr

1. How Russia blocks in 2026

TSPU is DPI hardware installed at every ISP (in-path, stateful), centrally controlled by Roskomnadzor. Unlike China's border-based GFW, it relies heavily on behaviour and IP reputation rather than protocol signatures alone.

flowchart TD
    C[Client in Russia] --> T{TSPU at ISP}
    T -->|1. Destination IP / ASN reputation| A1[Foreign datacenter ASN?
Hetzner, DO, OVH, Vultr...] T -->|2. TLS ClientHello scoring| A2[Fingerprint JA3/JA4
Chrome 133 / Safari / iOS flagged] T -->|3. Behaviour| A3[> 3 TLS handshakes to same SNI
within ~350-400 ms / 60 s] T -->|4. Data budget| A4[~15-20 KB / ~25 packets
from suspicious IP] T -->|5. Active probing| A5[Probe server; no real
website behind it = proxy] A1 & A2 & A3 --> F1[Silent freeze 120 s
no RST] A4 --> F2[TCP stalls after 16 KB] A5 --> F3[IP added to blocklist
whole /24 or ASN may follow] T -->|Mobile networks only| W[Whitelist mode:
default-deny, only approved IPs/SNIs]

Loading

Blocking timeline

timeline
    title Russian VPN blocking escalation
    Nov 2025 : First VLESS signature blocks
             : Connection-count policing on some home ISPs
    Feb 17 2026 : VLESS+REALITY over raw TCP hit on wired ISPs
                : ~16 KB randomness sampling
    May 21-25 2026 : Behavioural module rollout (Siberia, Moscow)
    Jun 5 2026 : Largest wave, cloud hosters hit
               : Fingerprint + burst freeze model
    Jul 2026 : Foreign-SIM roaming filtered like domestic
    Aug 2026 : Whole hoster subnets / ASNs blocked
             : 20+ VPN services down
    Sep 21-25 2026 : Post-election wave, subnet bans
    Oct 2026 : Default Xray chrome (Chrome 133) fingerprint targeted

Loading

Detection model (June 2026 reverse-engineering, "Siberian" model)

Signal Detail Penalty Confidence
Destination subnet / AS Foreign DCs long-listed; some Russian clouds (Selectel, Yandex Cloud, Cloud.ru) since late May Prerequisite Medium
TLS fingerprint chrome (uTLS Chrome 133), Safari, iOS flagged; Firefox, Edge, OkHttp lenient Prerequisite High (field tests Oct 13)
Handshake burst > 3 TLS attempts to one SNI, < ~350–400 ms apart within 60 s Freeze 120 s (was 600 s if fingerprint changed) Medium
Data budget ~15–20 KB server→client from suspicious DC IP TCP stall, no RST High
Active probing Static fallback, missing Server: header, 6-day certs = proxy IP blocklisted Medium
Subnet escalation One trigger can block suspicious subnets for minutes; whole ASNs since Aug Collateral damage High

2. Russia vs China — why Chinese defaults fail in Russia

Dimension 🇷🇺 Russia (TSPU) 🇨🇳 China (GFW)
Architecture DPI at every ISP, close to user Border gateways + regional (e.g. Henan)
Default stance Suspicious of foreign datacenter destinations; mobile whitelist = default deny Allow by default; block when protocol is identified
Main detection Burst count, fingerprint, data budget, subnet/ASN bans Active probing, fully-encrypted-traffic entropy rules, TLS-in-TLS ML, QUIC SNI
Block style Silent freeze, recovers after timeout RST injection, IP blackhole (rarely lifted)
Bottleneck The blocking itself Peak-hour international congestion / packet loss
Best primary XHTTP + XMUX (few connections) Vision over raw TCP
Fingerprint firefox / edge default chrome fine
SNI No big foreign brands; same-ASN neighbour or self-steal Same-ASN neighbour or own site

Key insight: Vision over raw TCP opens a new TLS connection per proxied connection — harmless in China, but exactly what TSPU's burst rule punishes. Since Sept 2025, VLESS Encryption lets Vision run over XHTTP, combining China's anti-TLS-in-TLS defence with Russia's need for few, multiplexed connections.


3. Protocol ranking

Rank Protocol Anti-DPI (RU) Speed Mobile Whitelist Setup Confidence
🥇 1 XHTTP + REALITY + VLESS Enc + Vision + XMUX (firefox) ★★★★★ ★★★★ ★★★★ needs relay Medium Medium-High
🥈 2 TCP + REALITY + Vision (firefox) ★★★ ★★★★★ ★★★ needs relay Low High
🥉 3 Hysteria2 (Salamander, port hopping, BBR) ★★★★ ★★★★★ on lossy links ★★ (operator-dependent) ❌ UDP blocked Low Medium
4 NaiveProxy (real Chromium stack) ★★★★ ★★★★ ★★★ ❌ High Medium
5 XHTTP via Russian CDN + VLESS Enc ★★★★ ★★ ★★★ ⚠️ if CDN whitelisted High Low-Medium
6 AmneziaWG 3.1 ★★★ ★★★★ ★★ ❌ Low Medium
7 MASQUE → Cloudflare WARP (Xray) ? ★★★ ★★★ (1 report) ❌ Low Low
8 Tor WebTunnel ★★★ ★ ★★★ ❌ Low Medium
— AnyTLS, TUIC v5, TrustTunnel, Mieru, Juicity ? ★★★★ ? ❌ Medium No Russian data
✗ ShadowTLS v3 ★★ (2 public detection vectors) — — — — Avoid
✗ Shadowsocks / Outline, WireGuard, OpenVPN, IKEv2, L2TP, SOCKS5, MTProto dead — — — — Avoid

Speed under packet loss (indicative, third-party test, 100 Mbit/s VPS)

Loss REALITY Hysteria2 Trojan
0 % 89 Mbit/s 92 Mbit/s 85 Mbit/s
5 % 42 Mbit/s (−53 %) 78 Mbit/s (−15 %) 35 Mbit/s (−59 %)
10 % ≈ −68 % ≈ −22 % ≈ −75 %
Vision has the best multi-threaded throughput; XHTTP has lower latency (multiplexed). No independent head-to-head benchmark exists — run your own.

4. Server location & routes

User location Primary exit Typical RTT Notes
St Petersburg / North-West Finland (Helsinki), Estonia, Latvia 15–25 ms Northern route via Stockholm/Helsinki
Moscow / Central Finland, Netherlands, Germany (Frankfurt) 30–50 ms Moscow–Minsk–Warsaw–Frankfurt ≈ 31 ms RTD
Siberia / Far East Europe (no good alternative found) higher New blocks are tested here first — keep backups

Hosting rules

✅ Do ❌ Don't
Small / niche European ASNs with no registry-listed ranges (AlphaVPS, UpCloud, HostHatch, Netcup, TransIP, Tilaa) Hetzner, DigitalOcean, OVH, Vultr, Contabo, Scaleway (on block lists); Clouvider, LiteServer, Worldstream, Keyweb (partly listed ASNs)
Two nodes in different countries and ASNs Stark Industries → MIRhosting → THE.Hosting chain (Dutch raids May–Jun 2026)
Ask hoster for a new IP in a different /24 when hard-blocked Russian VPS as exit (hoster bill, ID checks, 30-min block proposals)
Consider a residential exit abroad (not a DC ASN) Rotate IP for short flaps — it won't help

Host selection — live probes from inside Russia (2026-10-25)

Method.

  1. Live IP-block probes: 20 candidate IPs were traced through the cheburcheck.ru network of 32 probes inside Russia (Rostelecom, MTS, Beeline, Dom.ru, TTK and regional ISPs, Moscow to Yakutia), with 28 valid verdicts per IP.
  2. ASN sweep: 79 hosting ASNs were checked for any ranges already in Russian block registries.
  3. Terms check: 8 providers' official terms were read for VPN policy, Russia/KYC, extra IPv4 and refunds.
⚠️ This probe detects hard IP blocks only, not the 16 KB freeze — the Hetzner control (known to be frozen) passed 28/28. Test the freeze yourself after buying, inside the refund window.
flowchart LR
    A[89 'clean' hosts
not in registries] --> B{ASN has listed ranges?} B -->|Yes: Clouvider, LiteServer,
Worldstream, Keyweb, Webdock-HEL...| X[Deprioritise] B -->|No| C{Live probe
28 Russian vantage points} C -->|Any IP block| Y[Caution:
Creanova, Alwyzon, Webdock-DK, DEAC] C -->|28/28 OK| D{Terms + price} D --> W1[🥇 AlphaVPS Nuremberg] D --> W2[🥈 UpCloud Helsinki] D --> W3[🥉 HostHatch Stockholm/AMS]

Loading

Live probe results

Host Location OK / IP-blocked / unclear (32 probes) ASN has listed ranges? Verdict
AlphaVPS Nuremberg 28 / 0 / 4 No 🟢
AlphaVPS Sofia 28 / 0 / 4 No 🟢
UpCloud Helsinki 28 / 0 / 4 No 🟢
HostHatch Amsterdam / Stockholm 28 / 0 / 4 No 🟢
Netcup Germany 28 / 0 / 4 No 🟢
TransIP / Tilaa Netherlands 28 / 0 / 4 No 🟢
GleSYS Stockholm 28 / 0 / 4 No 🟢
Time4VPS / Cherry Servers Lithuania 28 / 0 / 4 No 🟢
Webdock Denmark 27 / 1 / 4 No (its Helsinki ASN: yes) 🟡
Alwyzon Vienna 27 / 1 / 4 No 🟡
Creanova Finland 26 / 2 / 4 No 🟡
DEAC Riga 26 / 2 / 4 (incl. Rostelecom) No 🟡
LiteServer Netherlands 27 / 1 / 4 Yes 🟠
Clouvider Amsterdam 26 / 2 / 4 Yes 🟠
Hetzner (control) Helsinki 28 / 0 / 4 Yes 🔴 known 16 KB freeze

ASNs with ranges already in registries (9/79): Clouvider, LiteServer (Infrastructure Group), Worldstream, Keyweb, PQ.Hosting/Aeza, RackNerd (HostPapa), RamNode (Hivelocity), WaveCom, Webdock-Helsinki (Data Room).

Terms of service (verified on official sites)

Provider Personal VPN Russia / KYC Extra IPv4 Refund
AlphaVPS ✅ explicitly allowed ("only used by the customer") No Russia clause; anonymous orders rejected, KYC possible not found ✅ 14 days, first order, unused
UpCloud not mentioned Sanctions list names Cuba, Iran, NK, Syria, Crimea — not Russia €3.47/mo ❌ none, but hourly billing
HostHatch not mentioned not found ~$2/mo (API sample) ✅ 7 days, reason required
Netcup not mentioned ID + selfie verification or prepay €2/mo statutory 14 days only
Webdock ✅ markets to VPN providers may verify identity €1.75/mo ❌ none
Creanova not mentioned real postal address €1.40/mo not found
Alwyzon Tor banned; VPN not mentioned sanctions warranty €2/mo + €5 setup statutory 14 days only
Clouvider commercial/public VPN banned ❌ refuses Russian residents ~£2/mo ❌ none

Final ranking

Rank Host Price Specs Why
🥇 AlphaVPS Nuremberg (C2G) €2.99/mo 2 vCPU / 2 GB / 30 GB / 3 TB / up to 10 Gbit/s 28/28 probes, clean ASN, personal VPN explicitly allowed, 14-day refund, crypto accepted
🥈 UpCloud Helsinki (Starter) €3/mo 1 vCPU / 1 GB, IPv4 included 28/28 probes, clean ASN, mainstream cloud; hourly billing = delete & redeploy for a new IP in minutes
🥉 HostHatch Stockholm / Amsterdam $4/mo 1 vCPU / 2 GB / 1 TB 28/28 probes, 7-day refund
4 Netcup €9.50/mo 2 vCPU / 4 GB 28/28 probes; ID + selfie KYC
— Clouvider £3.50/mo 1 vCPU / 2 GB / 10 Gbit/s ❌ dropped: listed ranges, 2 probe blocks, refuses Russian residents, no refunds

Recommended pair: AlphaVPS Nuremberg (primary) + UpCloud Helsinki (backup) ≈ €6/month — different countries, different ASNs, one refundable, one with instant IP rotation.

After buying: inside AlphaVPS's 14-day window, test the 16 KB freeze from Russia (hosting-checker / dpi-checkers) and run the primary profile for 2–3 minutes. Refund and switch to HostHatch if it fails. On UpCloud, redeploy for a new IP if needed.


5. Best deployment

Architecture

flowchart LR
    subgraph RU[Russia]
        H[Home broadband / Wi-Fi]
        M[Mobile, normal mode]
        MW[Mobile, whitelist mode]
        RL[Russian relay
whitelisted subnet
Yandex Cloud / VK Cloud / Timeweb] VC[VK Calls / Telemost
WebRTC tunnel] DNS[ISP / NSDI resolver
195.208.4.1, Yandex DoT] end subgraph FI[Server A — AlphaVPS Nuremberg] P1[① XHTTP + REALITY
+ VLESS Enc + Vision
XMUX=2, :443] P2[② TCP + REALITY + Vision] end subgraph NL[Server B — UpCloud Helsinki] P3[③ Hysteria2
Salamander + port hop] P4[④ NaiveProxy or AmneziaWG 3.1] end H --> P1 H -.speed.-> P2 M --> P1 H -.fallback.-> P3 M -.fallback.-> P4 MW --> RL -->|XHTTP + VLESS Enc| P1 MW -.emergency.-> VC --> P2 MW -.last resort.-> DNS -->|dnstt / slipstream| NL

Loading

Failover decision flow

flowchart TD
    S([Connect]) --> Q1{Mobile whitelist active?}
    Q1 -->|No| A[① XHTTP + REALITY + Enc + Vision]
    A --> Q2{Works for 2-3 min?}
    Q2 -->|Yes| OK([Use it; Vision TCP for big downloads at home])
    Q2 -->|Freeze| W[Wait 130 s, do NOT change fingerprint]
    W --> Q3{Plain HTTP to server works?}
    Q3 -->|Yes - TLS scoring| B[Switch SNI / server; check XMUX connections]
    Q3 -->|No - IP blocked| C[③ Hysteria2 / ④ on Server B
request new IP in another /24] Q1 -->|Yes| D[Russian whitelisted relay → Server A] D --> Q4{Relay alive?} Q4 -->|No| E[WebRTC video-call tunnel] E --> Q5{Works?} Q5 -->|No| F[DNS tunnel via ISP/NSDI resolver] Q5 -->|Yes| OK Q4 -->|Yes| OK

Loading

Primary client outbound (Xray JSON)

{
  "protocol": "vless",
  "settings": { "vnext": [{ "address": "server-a.example", "port": 443, "users": [{
    "id": "UUID", "flow": "xtls-rprx-vision",
    "encryption": "" }] }] },
  "streamSettings": {
    "network": "xhttp",
    "xhttpSettings": { "path": "/random-path", "mode": "stream-one",
      "extra": { "xmux": { "maxConnections": 2, "cMaxReuseTimes": 0,
        "hMaxRequestTimes": "600-900", "hMaxReusableSecs": "1800-3000" } } },
    "security": "reality",
    "realitySettings": { "serverName": "same-ASN-neighbour.example",
      "fingerprint": "firefox", "publicKey": "PUBLIC_KEY", "shortId": "SHORT_ID" }
  }
}

Server checklist

Item Setting
Xray version ≥ v26.10.21 (security update); avoid v26.7.11 (REALITY/Shadowrocket regression)
REALITY target Same-ASN neighbour found with RealiTLScanner, or self-steal (own domain + real nginx site + Let's Encrypt)
SNI naming Never vpn, proxy, node, xray, reality; never google/microsoft/apple
decryption Server string from xray vlessenc; no fallbacks (incompatible)
Other ports Forward TCP/80 and UDP/443 to target; SSH on high port, IP-restricted
Fallback rate limits Off, or randomized (fixed values are a fingerprint)
XMUX Set all fields explicitly (setting one clears defaults); verify with ss -tn
Users Personal / family only — more users = more visible profile

Clients

Use Client Why
Primary (VLESS Enc + XHTTP) Happ (Xray ≥ 26.9), v2rayNG, v2rayN Xray-core; upstream sing-box (Hiddify, NekoBox) lacks VLESS Encryption
Backups (Hysteria2, AWG 3.1) mihomo ≥ 1.19.30 clients Supports Hysteria2, AmneziaWG 3.0/3.1, VLESS Encryption
Whitelist emergency kulikov0/whitelist-bypass, SlipNet / dnstt clients WebRTC / DNS carriers

6. Mobile whitelist mode

Since 2025–2026, many regions switch mobile internet to a default-deny whitelist (≈ 0.14 % of Russian IPv4 addresses reachable in one MegaFon scan). Per the Ministry of Digital Development (2026-03-21) this applies to mobile only — wired / Wi-Fi is unaffected.

Channel Mechanism Measured speed Status Source
Russian whitelisted relay → foreign exit Entry IP inside whitelisted /24 26 Mbit/s single, 94 Mbit/s multi, +34 ms Best; IPs hard to get; ministry pressure Friskes/vpn-infra (Aug 2026)
WebRTC video-call tunnel Data inside VK Calls / Telemost / WB Stream SFU ~6.5 Mbit/s per track (author) Active project, account risk net4people #618
DNS tunnel (dnstt / slipstream) Queries via ISP or NSDI resolver 0.3 KB/s – 3 Mbit/s Operator-dependent, degrades over time Meduza, vpn-infra
Yandex Cloud Functions bridge Serverless inside whitelist Medium Needs verified account Habr
VK TURN relay WireGuard via TURN ~250 kbit/s Throttled Habr
Foreign SIM roaming Home-routed traffic — ❌ filtered since Jul 2026; 24 h cooldown te-st.org
2G / EDGE Unfiltered at many sites very slow Text-only emergency vpn-infra

7. Lessons borrowed from abroad

Source Technique Applies to Russia?
🇨🇳 China REALITY, Vision, RealiTLScanner, XHTTP up/down split ✅ core of the recommendation
🇨🇳 GFW Report (USENIX'23) Fully-encrypted-traffic exemption rules → Xray finalmask Sudoku / header-custom ⚠️ promising for non-TLS fallback, untested in RU
🇨🇳 GFW Report (PETS'26) CensorLess rotating serverless bridges (~$0.27/month) ✅ fits whitelist mode (Yandex Functions)
🇮🇷 Iran SNI-whitelist phases, fake ClientHello desync (lasted ~4 days), DNS tunnels, WebRTC carriers ✅ preview of Russia's next steps — prepare diverse backups
🇮🇷 Iran Starlink ❌ not licensed in RU, legal & physical risk
Tor / Psiphon WebTunnel; Conduit volunteer residential proxies ⚠️ WebTunnel recommended for RU mobile; Snowflake DTLS targeted since Mar 2026

8. Operations playbook

  1. One subscription link with all profiles; client url-test / fallback groups.
  2. Test each profile 2–3 minutes, waiting 130 s and toggling airplane mode between runs. Tools: hyperion-cs/dpi-checkers, siberian-check, hosting-checker.
  3. Test on the major home ISPs (Rostelecom, MTS/MGTS, ER-Telecom/Dom.ru, Beeline) and 2–3 mobile operators.
  4. Never change fingerprint during a freeze — wait or switch node.
  5. Distinguish failures: plain HTTP works but TLS fails → behavioural/fingerprint; nothing answers → IP blocklisted (check from abroad via check-host.net).
  6. Update Xray weekly; rotate between 2–3 SNIs; keep a spare IP + redeploy script.
  7. Expect tightening around elections and major events.

9. Confidence & open questions

Level Claims
High Vision+REALITY fundamentals; burst rule exists; Chrome 133 fingerprint targeted; SS/WG/OpenVPN dead; whitelist = mobile only
Medium Fused XHTTP + Enc + Vision effectiveness; Hysteria2 on mobile; AmneziaWG 3.1; exact thresholds
Low MASQUE/WARP, TrustTunnel, AnyTLS in RU; whitelist bypass speeds (small samples); finalmask Sudoku vs TSPU; residential exits avoiding the data budget

Sources

Xray / GitHub

net4people & community

Analysis & reporting

Academic


This report is for research and digital-rights purposes. Laws on VPN use and hosting vary; assess your own legal and personal risk.

参与讨论