Hysteria2 is worth testing on networks that allow UDP, especially links with high round-trip times or random packet loss, but it is not guaranteed to outperform other transports. This guide is for Shadowrocket users who have obtained configuration details from their own service provider. It explains QUIC and congestion control, then covers parameter checks, Global Routing tests, and troubleshooting. Shadowrocket is available for iPhone, iPad, Mac, Windows, Android, and Linux. The steps below refer to Home, Config, and Settings on mobile; on other platforms, check the corresponding fields.
How Hysteria2 transports data
Hysteria2 is a proxy protocol built on QUIC. To understand how it works, distinguish between the application request, the proxy session, and the underlying packets. A browser may send a TCP request, while the connection between Shadowrocket and the server uses QUIC over UDP. The destination website does not need to run Hysteria2 or know which proxy transport the client uses.
UDP alone does not guarantee delivery, ordering, or retransmission. QUIC adds connection management, encryption, reliable streams, and loss recovery on top of it. Seeing UDP does not mean web data can be lost arbitrarily. QUIC handles retransmission and ordering for data that requires reliable delivery; forwarding UDP through a proxy does not turn all traffic into reliable streams.
The same port number does not mean the same service. A network allowing 443/TCP for websites does not necessarily allow 443/UDP as well. Changing the port in a configuration to 443 will not make a service available if it is not listening on that port. Use the address, port, and authentication details supplied by your service provider.
Why it can be faster
Throughput on a high-latency link depends on more than bandwidth; it also depends on whether the sender can keep enough data in flight. For example, at 100 Mbps with a 150 ms round-trip time, the bandwidth-delay product is about 1.875 MB. That is roughly the amount of data in flight needed to sustain that rate—not a speed-test result or a value to enter in the client.
When a transport algorithm treats random packet loss as congestion, it may reduce its sending rate. Waiting for acknowledgments before ramping back up also takes longer when round-trip times are high. Hysteria2's congestion-control design offers a bandwidth-informed sending strategy that may sustain throughput better than more conservative approaches on some high-latency, randomly lossy links. The control method actually used also depends on client settings, server configuration, and negotiation between the two.
- Stream independence: QUIC streams do not all have to wait for missing bytes from one stream before proceeding, but they still share link capacity and connection-level resources.
- Loss recovery: Packet loss still has to be detected and recovered from. QUIC cannot eliminate wireless interference or create more bandwidth.
- Bandwidth estimates: If the configuration asks for upload and download bandwidth, enter sustainable rates—not the advertised speed of your broadband plan as though it were a live measurement.
- Congestion limits: Continuing to increase the sending rate during real congestion can worsen queueing. Speed-test numbers may rise even as interactive response times get worse.
Bottom line: track throughput and responsiveness
When evaluating performance, track sustained download speed, time to first page render, and responsiveness during downloads. Keep the current settings only if throughput improves while everyday response times remain acceptable.
Which network conditions suit Hysteria2?
Hysteria2 is most worth testing on long-distance, high-latency links with non-congestion-related packet loss. Performance on mobile or Wi-Fi networks depends on carrier policies, signal quality, and UDP reachability; using QUIC alone does not guarantee a more stable connection. On low-latency, low-loss networks with plenty of bandwidth, differences between protocols may be smaller than differences in server load.
| Network symptom | What it may indicate | What to try |
|---|---|---|
| High round-trip time; throughput drops during sustained transfers | Possible window-growth or loss-recovery overhead | Compare sustained throughput and responsiveness against the same target at the same time of day |
| Websites connect directly, but Hysteria2 keeps timing out | Possible UDP path, server listener, or configuration issue | Try again on Wi-Fi or mobile data, then check the address and port |
| Everything slows down during evening congestion | More likely limited link capacity or server resources | Lower the bandwidth target to avoid increasing queueing by sending more data |
| Connection succeeds; small requests work, but large transfers stall | Possible path MTU, packet loss, or middlebox handling issue | Keep the logs and investigate one cause at a time; do not start by applying an arbitrary MTU value |
Distinguish between the server's ability to forward UDP to a destination and the current network's ability to reach the server over UDP. The former is a proxy capability; the latter is a matter of network reachability. If the network cannot reach the server over UDP, the Hysteria2 session itself may not establish, and changing routing rules usually will not help.
Check the settings in the client
Get your subscription link from your service provider. If you already have a subscription, keep the complete imported configuration rather than copying only the address and port and entering them manually. You cannot convert an older protocol entry to Hysteria2 just by changing its display name; make sure the Type matches the protocol actually running on the server.
-
Check the protocol
Open Home and edit an existing entry. To add one manually, go to Home → + → Type and select Hysteria2. If this type is not listed in your interface, check the download guide for supported platforms and installation options. Do not substitute a similarly named type.
-
Check the address
Check Address and Port. Enter the server hostname or IP in Address and the port number supplied by your service provider in Port. Do not paste an entire subscription URL containing a path or query parameters into the server address field.
-
Check authentication
Check Password and the TLS settings. SNI must match the server certificate and connection configuration; do not enter an arbitrary domain. Do not disable certificate verification to hide a certificate error.
-
Check additional settings
If obfuscation is enabled on the server, the client must use the matching obfuscation method and password. If the interface provides bandwidth fields, check the units and any auto-negotiation notes first. Mbps and MB/s differ by a factor of eight; do not confuse them.
-
Save and test
Save, return to Home, select the entry, check Global Routing, and connect. Test a website you know is reachable, then try a large file transfer. Do not judge whether the protocol works based on a single latency result.
Choose a routing mode
The protocol determines how the client and server communicate; Global Routing determines whether a request takes the proxy path. These are separate settings. If a request matches DIRECT in Config, it does not use the selected Hysteria2 entry, so its speed cannot be used to evaluate Hysteria2.
Config
RecommendedRouting follows the rules in the currently enabled configuration. Rules are usually evaluated from top to bottom, with the first match taking effect. FINAL is the fallback when no earlier rule matches; it does not necessarily mean direct connection.
Best for everyday use and networks where you need to distinguish clearly between proxy and direct connections.
Proxy
Route requests handled by the client through the proxy. This can help rule out interference from unmatched rules. Still, check the selected entry and policy group; this does not mean all system traffic on the device is being routed through the proxy.
Best for briefly checking a Hysteria2 connection and comparing the effect of routing rules.
Direct
Use a direct connection without the selected proxy server. This can help confirm that the underlying network works, but it does not test Hysteria2's transport performance.
Best for establishing a baseline and checking whether the issue is on the proxy path.
Scene in Global Routing applies settings for a given scenario; it is not a transport protocol. During troubleshooting, also check Settings → On Demand to prevent automatic connection behavior from interrupting manual tests. Restore your usual settings when testing is complete. Check the selected rules in Config separately from the selected entry in Home.
Troubleshooting connection failures
A normal latency test with websites still failing does not automatically point to Hysteria2. The test target and method may differ from a real request: reaching the server does not mean the destination domain resolved successfully or that the browser request matched a proxy rule. Troubleshooting by stage is more effective than changing several settings at once.
- Session will not establish: Check the system time, Address, Port, Type, and Password. Confirm that the server is listening on the specified UDP port, and check firewall rules and network access restrictions.
- TLS errors: Check SNI, the certificate's validity period, and its name. Correct the connection details first; do not make certificate verification bypass a routine setting.
- Fails on only one network: On the same device, switch between Wi-Fi and mobile data without changing the other settings. If the results vary by network, investigate UDP reachability, NAT behavior, and network restrictions first.
- Connected, but domain requests fail: Check DNS results, the active Config, and the matching policy. Distinguish resolution failures, destination connection refusals, and proxy timeouts instead of labeling them all as a protocol problem.
- Connection drops after running for a while: Check whether it happens only after idle periods or during sustained transfers too. The former may involve NAT mapping timeouts; the latter also calls for checking packet loss, bandwidth settings, and server load.
Change one setting at a time and record the symptoms before and after. If a subscription update overwrites manual changes, first check whether the server configuration has changed. Do not keep restoring old ports or authentication details. If you need help, share redacted error logs and steps to reproduce the issue—not your full subscription link.
Set up a reproducible test
The goal is to find out whether a particular combination of device, network, and server works well—not to rank protocols universally. Keep the device, network connection, target file, Global Routing settings, and test duration consistent. Server location, egress capacity, and load can outweigh differences between transport protocols.
- Record the test conditions: Note whether you used Wi-Fi or mobile data, the time of the test, the selected entry, and the routing mode. Do not include authentication details.
- Repeat measurements: For example, run a 60-second test three times and compare the steady-state results rather than a brief peak. This is a testing suggestion, not a performance guarantee.
- Check concurrent use: While a download is running, open a frequently used website and note any significant queueing, delay before the page appears, or reconnections.
- Control variables: Change only one bandwidth setting at a time and keep its original value. When finished, restore your usual Config and On Demand settings.
Bottom line: confirm reachability before tuning speed
If the connection is not stable yet, fix UDP reachability, authentication, and TLS settings first. Compare bandwidth settings and congestion control only after the connection is consistently usable.
Hysteria2 offers another transport option for specific network conditions; it is not a substitute for diagnosing DNS, routing rules, or the underlying network. Understand the limits of QUIC reliable streams, UDP reachability, and bandwidth targets to identify where the slowdown occurs and what to check next.