Hysteria2 Explained: A Fast VPN Protocol for China
Hysteria2 is a QUIC protocol built for lossy, throttled links. Why it holds up during China's peak-hour packet loss when TCP-based tunnels collapse.
There is a specific kind of bad evening in mainland China. The VPN is connected. A web page eventually loads. But the video call keeps freezing, the upload crawls and resets, and everything feels like it is moving through mud. The connection is not blocked. It is drowning in packet loss, and most protocols respond to packet loss by giving up speed. Hysteria2 is built to do the opposite.
Hysteria2 is a modern proxy protocol designed for exactly the networks that break ordinary VPNs: lossy, congested, throttled, high-latency links, which is a precise description of the cross-border path out of China during peak hours. It runs on a different foundation from most tunnels, and it uses an aggressive approach to congestion that keeps data moving when a normal connection would collapse. This guide explains what Hysteria2 is, why packet loss matters so much in China, how its congestion control differs from everything built on TCP, and where it fits in a setup that actually survives the evening rush.
Table of contents
- Why packet loss, not blocking, breaks the evening
- What Hysteria2 is built on
- Brutal congestion control, explained simply
- TCP versus Hysteria2 under packet loss
- What Hysteria2 does not solve on its own
- Where Hysteria2 fits in the China stack
Why packet loss, not blocking, breaks the evening
People assume the Great Firewall works by blocking. A lot of the daily pain is subtler than that. Instead of cleanly cutting a connection, the cross-border path degrades: some packets get through, some are dropped, latency swings, and the link becomes unreliable rather than dead. This is sometimes deliberate throttling and sometimes just raw congestion on the narrow set of international gateways, and the symptoms look the same from a user's seat.
The reason this is so destructive is the way most connections react to loss. The entire internet is built on TCP, and TCP treats packet loss as a signal that the network is overloaded. When it sees loss, it slows down, on purpose, to be polite and avoid making congestion worse. That is the correct behaviour on a healthy network. On a network where loss is constant, it is a disaster. A TCP connection on a link with 5 to 10% loss does not slow down a little. Its usable throughput collapses, often to a fraction of what the line could actually carry.
The trap: on a lossy China link, a normal connection assumes the loss means congestion and throttles itself. It is being polite to a network that is interfering with it on purpose. Politeness is exactly the wrong response.
This is why a connection can show plenty of raw bandwidth on paper and still fail a video call. The protocol is leaving most of the line unused because it is reacting to loss as if it were its own fault.
What Hysteria2 is built on
Hysteria2 starts from a different foundation. Three technical choices define it.
It runs on QUIC over UDP. Instead of TCP, Hysteria2 uses a customized version of QUIC, the modern transport that also powers HTTP/3. QUIC runs on UDP, which gives the protocol full control over how it handles loss and retransmission rather than inheriting TCP's cautious behaviour. It also means a single dropped packet does not stall an entire stream the way it can on TCP.
It masquerades as HTTP/3. Because real web traffic increasingly uses HTTP/3 (which is QUIC), Hysteria2 traffic can blend in with ordinary encrypted web browsing. To a censor, blocking it broadly risks collateral damage to legitimate HTTP/3 sites, which raises the cost of interference.
It is built for unreliable networks. This is the stated design goal of the project: unparalleled performance over lossy and unstable links. Everything about it is tuned for the conditions that censored and congested networks create, rather than for a clean data-center benchmark.
Brutal congestion control, explained simply
The feature that makes Hysteria2 stand out in China is its congestion control, associated with an approach called Brutal. The idea is almost rude, which is the point.
With a normal protocol, the connection constantly guesses how much bandwidth is available by watching for loss and backing off when it appears. Brutal flips that. You tell it how much bandwidth you actually have, and it sends at that rate, treating packet loss as something to overcome with retransmission rather than a reason to slow down. It keeps the pipe full instead of repeatedly retreating.
On a clean network this would be unnecessary and even antisocial, because it does not back off to share capacity fairly. On a deliberately degraded or congested cross-border link, it is exactly what keeps a video call alive. The connection pushes its data through the loss instead of surrendering throughput to it.
There are honest trade-offs. Brutal-style control is aggressive and not "fair" to other traffic in the way TCP is, and it depends on setting the bandwidth figure sensibly. Set it far above what the line can carry and you simply create more loss. Used correctly, though, it is the difference between a connection that holds 100 Mbps through 8% loss and one that collapses to a crawl.
TCP versus Hysteria2 under packet loss
The contrast is easiest to see side by side. This is the behaviour that matters during the 7 p.m. to midnight window in China.
| Condition | TCP-based tunnel (OpenVPN, WireGuard, most proxies) | Hysteria2 (QUIC, Brutal) |
|---|---|---|
| Clean network, low loss | Fast and efficient | Fast, no real advantage |
| Moderate loss (3 to 8%) | Throughput drops sharply as it backs off | Holds throughput by pushing through loss |
| Heavy peak-hour loss and throttling | Often collapses to near-unusable | Degrades gradually, stays usable for calls |
| Single dropped packet | Can stall the whole stream (head-of-line blocking) | Affects only the affected data, stream continues |
| Looks like | A recognizable VPN or a TCP-in-TCP pattern | Ordinary HTTP/3 web traffic |
The takeaway is not that Hysteria2 is universally faster. On a healthy link it offers little over a good TCP setup. Its advantage shows up precisely when the network is bad, which in mainland China is most evenings. That is why it is valued as a peak-hour and lossy-link protocol rather than an everyday default for every situation.
What Hysteria2 does not solve on its own
Hysteria2 is powerful, but it is one layer, and a few realities keep it honest.
- It does not hide as deeply as a borrowed-handshake protocol. Its HTTP/3 masquerade is good, but it relies on UDP, and some networks throttle or restrict UDP, or treat unusual QUIC traffic with suspicion. For pure stealth against active probing, a TLS-borrowing protocol like VLESS-Reality is harder to detect. The two are complementary, not interchangeable.
- It does not replace a good route. Brutal can push through loss, but it cannot create capacity that the international gateway does not have. On a premium route there is simply less loss to fight in the first place, which is the case for running it over CN2 GIA rather than commodity transit.
- It needs correct settings. The bandwidth value has to reflect reality. A consumer who self-hosts can get this wrong and make things worse. This is part of why a managed setup tends to beat a do-it-yourself one, a theme throughout our guide on whether VPNs actually work in China.
Where Hysteria2 fits in the China stack
The strongest China setups do not pick one protocol and hope. They layer protocols so each covers the other's weakness, and Hysteria2's role is specific.
- CN2 GIA is the premium route, so there is less congestion and loss to begin with.
- VLESS-Reality is the primary transport for everyday use, because it is the hardest to detect and survives active probing.
- Hysteria2 is the fallback for bad evenings, because when loss spikes and a TCP-based tunnel would stall, its QUIC transport and aggressive congestion control keep calls and uploads alive.
Think of VLESS-Reality as the protocol that wins the stealth fight and Hysteria2 as the protocol that wins the packet-loss fight. A serious service runs both and switches based on conditions, rather than forcing one protocol to do a job it was not designed for. That layered approach is the practical difference our ranked guide to the best VPN for China keeps pointing to, and it is why peak-hour behaviour, not a noon speed test, is the real measure.
Hysteria2 is the protocol that refuses to slow down when the network tries to make it. On a lossy China evening, that stubbornness is the feature.
Throughwire uses Hysteria2 as a peak-hour fallback alongside VLESS-Reality, both running over CN2 GIA premium routing, so the connection stays usable even when the cross-border path is at its worst. For travelers, expatriates, and teams who need calls and uploads to survive the 9 p.m. congestion window inside mainland China, Throughwire is built on exactly this layered stack.