If a speed test shows a latency of just a few dozen milliseconds but webpages still take ages to load, don’t switch protocols yet. Test request time and download throughput separately against the same target, then compare nodes, network connections, and settings one at a time. Use v2rayN on desktop. On Android, use v2rayNG with the Xray core or v2flyNG with the v2fly core. Your results should tell you whether everything is slow or the problem is limited to a particular node, network, or app.
Set a baseline: measure request time and download speed separately
For desktop, this example uses v2rayN 7.x: right-click a node in the list and select “Test server real connection latency.” On Android, find the connection test available in your version of v2rayNG or v2flyNG, then retest in the affected app. The client test uses its configured test address, while the app follows a different end-to-end path. One test cannot replace the other.
- Pause cloud sync and background downloads. Record the node, network, proxy mode, test target, and time. Run each test three times and note the median, along with any timeouts. Don’t count timed-out samples as zero milliseconds.
- Choose a fixed file that can be downloaded repeatedly. Use the same app to download the same file, and make sure you’re seeing a network response rather than a cached copy. Record the time until data starts arriving and the sustained speed after the download begins. Use a smaller sample on metered mobile data.
- Change just one thing per round. For example, switch from node A to B on the same Wi-Fi without changing DNS or proxy mode. Test node A again afterward to check whether the difference was just a short-term network fluctuation.
First, find out where the time is going
If all three runs take a long time to receive the first byte but download steadily once they start, keep the current node and check DNS resolution, connection setup, and the target site’s response first. If the first byte arrives promptly but throughput remains low, compare nodes and network routes.
Node load: compare two nodes on the same network
Choose nodes A and B in the same region with similar transport settings, then connect in the order A, B, A. Start a fresh request after each switch; don’t reuse a long-lived connection established through the previous node. Test the same download target at roughly the same time. If only one target is slow, try another to avoid mistaking the site’s rate limit for an overloaded node.
- Only A slows down in the evening: Add an off-peak comparison. If the pattern repeats, switch to B and note when A slows down. Node load, rate limits, and congested exits can all produce this result; a client speed-test number alone can’t prove the server is overloaded.
- A and B both slow down: They may share an exit, or both may be affected by your local network. If possible, test a third node known to use a different route, then compare another network connection.
- It slows down only when multiple tasks run: Keep one download running and pause uploads on other devices. When upload capacity is saturated, requests and acknowledgments can queue up too. If the slowdown returns when concurrent tasks resume, a single low-latency test while the network was idle doesn’t rule it out.
Connection quality: keep the node fixed and change networks
Use the same device, node, and test target. Test over Wi-Fi, then switch to a mobile hotspot or another available network. After switching, confirm the app is using the new connection and start a fresh request. Don’t change networks and node regions at the same time, or you won’t know whether the improvement came from local access or the remote route.
| What you observe | Next step |
|---|---|
| Occasional timeouts on Wi-Fi; all three hotspot tests succeed | Switch back to Wi-Fi and retest near the access point. If your desktop supports Ethernet, compare wired and wireless connections. |
| Node A is slow on both networks | Investigate node A’s address, transport settings, and exit route first. Don’t rush to change local DNS. |
| The same site is slow on both networks | Test a second target and confirm that requests to the original site match the expected routing rule. |
If available, run a brief network diagnostic against the same host. Note that a regular ping tests ICMP round-trip time and doesn’t use the app’s proxy connection. Some hosts don’t respond to or limit these probes. A ping timeout alone doesn’t prove packet loss on the proxy route; consider it alongside real connection tests, interrupted downloads, and log timestamps.
Transport settings: use logs to distinguish timeouts, refusals, and certificate errors
If failures start right after updating your subscription, compare the node details before and after the update. In v2rayN, open the node’s edit window and check its address, port, protocol, transport, and security settings. For TLS nodes, also check the server name. For WebSocket nodes, verify the path and Host value supplied by your provider. HTTPS commonly uses port 443, but use the port specified in your node configuration. Don’t switch VMess, VLESS, or transport types just to improve speed; the server must support the selected combination.
Error: i/o timeout
Cause: A network operation exceeded its timeout. This message alone doesn’t tell you which hop failed. What to do: Check the address and prefix on the same log line. First determine whether the delay occurred during DNS resolution, connection to the node, or the handshake, then retry the same operation on another network.
Error: connection refused
Cause: The connection was refused, often because no service is listening on the target port. What to do: Check whether the log shows a local or remote address. For a local address, make sure the core is running and listening on that port. For a remote node, verify the port in your subscription and ask your provider to confirm the service is up.
Error: x509: certificate signed by unknown authority
Cause: The current core doesn’t trust the certificate chain it received. What to do: Verify the node address and TLS server name, confirm you’re connecting to the expected service, and ask your provider to check the certificate chain. Don’t hide the error by disabling certificate verification.
Capture the logs around one failed request. Before sending them to the maintainer, remove subscription URLs, UUIDs, access tokens, and unrelated request records. Keep the timestamps, failure stage, port, and key error messages. If multiple nodes in the subscription fail, mention whether this started right after the nodes were updated.
Local proxy and DNS: confirm how app traffic is routed
A successful client test only confirms that its test request works. On Windows, check v2rayN’s local listener settings under “Settings” → “Parameter Settings,” then verify the current system proxy status from the tray menu. If an app requires manual proxy settings, enter the protocol and port shown by the client; don’t use a SOCKS port as an HTTP port. The system proxy may not cover every app. Open the same target in two apps first to see whether only one is bypassing the system settings.
- System proxy vs. TUN: If your current version supports both, test one traffic-routing method per round. Note the original settings before switching to TUN, and avoid running it alongside other tools that intercept traffic. If only TUN is slow, check the logs and routes to confirm requests use the expected exit before switching nodes.
- Android app coverage: Check that the VPN is connected in v2rayNG or v2flyNG, then see whether app filtering is enabled. Add the affected app to the proxy scope and retry. If only one app is affected, check its own proxy and DNS settings too.
- Compare DNS behavior: Resolving the node address and resolving the target domain are two separate operations. Note whether the first request stalls during resolution, then check the logs for the failing domain. Retry with the same domain, then compare with a second one. First identify whether the system, app, or core is making the query; change only the relevant setting. Replacing an HTTPS hostname with an IP changes certificate and site-matching behavior, so it isn’t an equivalent test.
FAQ: choose your next test based on the symptoms
Why does video still stutter when the speed test shows only 60 ms?
Connection-test latency doesn’t indicate sustained throughput. Keep the same node and play the same video at the same quality twice, then download a fixed test file. If only video stutters, check routing for the video site or its media requests. If downloads are slow too, compare nodes A and B.
My phone is fine but my computer is slow. Should I update my subscription first?
Connect both devices to the same Wi-Fi and test the same node and target. Make sure the phone hasn’t switched to mobile data. If only the computer is still slow, check the Windows system proxy, the app’s own proxy settings, and background uploads before changing your subscription.
After updating my subscription, how can I check that I’m still using the same route?
Don’t rely on the node’s display name alone. Open the updated node and compare its address, port, protocol, transport, and security settings. If they differ, test the new configuration three times on its own and record the results. Don’t use the old latency as a measure of the new route.
DNS was faster once after I changed it. Should I keep the new setting?
Test the original and new settings three times each, keeping the target, node, and proxy mode fixed. Record the first request separately from subsequent requests. Keep the change only if it consistently fixes the corresponding DNS issue; one faster download isn’t enough to attribute the improvement to DNS.