For v2rayN users who can connect but experience slow page loads, sluggish downloads, or transfers that stall. Establish a baseline using the same test target, then change nodes and access networks one at a time. Check routing, DNS, and TUN MTU first only if results change with the local mode. Change one setting per round, and keep a record of the original value and each retest.
Low latency doesn't mean fast downloads
A TCP probe mainly measures how long it takes to connect to the probe endpoint. An HTTP latency test through a proxy may also include the proxy handshake and the test site's response time. Both tests typically exchange very little data, so they don't show how much throughput a long-running connection can sustain. Speed tests have limits, too: results from one download URL don't represent every website.
Check the local entry point
In the v2rayN main window, open Settings → Parameter settings and check the current local listening port. At the bottom, note the selected System proxy and Routing options, and whether TUN is enabled. Menu locations may vary by version; use the port from the configuration currently in use.
Use a fixed download target
Choose a fixed HTTPS file you have permission to access, and download it with the same tool each time. Turn off parallel downloads and background sync. Make sure requests go through the client and that each run fetches data from the network rather than the browser cache. When comparing proxy and direct connections, use a target accessible both ways.
Record timing separately
Record time to first byte, total time, and any stalls. For example, a 100 MiB file downloaded in 25 seconds averages 4 MiB/s end to end. If the first 8 seconds are spent waiting for the first byte, the average transfer speed for the remaining time is about 5.9 MiB/s. This is a calculation example, not measured node performance.
Repeat the comparison
Run each condition three times in a row, then reverse the order and repeat—for example, test A then B, followed by B then A. Along with the median, record the lowest speed and number of stalls. A single slow outlier isn't enough to tell network jitter from temporary throttling by the target service.
Layer 1: Node load and exit path
Keep the access network, client mode, and download target fixed while alternating between two nodes. If A is repeatedly slow and B repeatedly fast, focus first on the path used by A. Switching nodes changes the server, entry point, and exit path, so the result alone doesn't prove that A's CPU is saturated.
- Test two targets on the same node: If both independent targets are slow, check the node's shared network path first. If only one is slow, consider target-side throttling, differences in content delivery, or the route from the exit to that target. Two URLs hosted by the same download service aren't independent comparisons.
- Repeat the same tests at different times: For example, run one round at 2:00 p.m. and another at 9:00 p.m. If speeds drop only in the evening, the issue is time-dependent. Shared node bandwidth and access-network congestion can both vary by time of day, so compare another network as well.
- If you have server access, check the same time window: Match the period of slow client speeds against server CPU, outbound bandwidth, and egress errors. Suspect saturated outbound capacity only when usage stays near a known bandwidth limit and requests from multiple sources slow at the same time. Without monitoring, share the node name, time, and retest results with the service provider. Don't forward subscription links or user identifiers.
Layer 2: Access network and transmission path
Keep the node and download URL fixed. Try Ethernet instead of Wi-Fi, or switch to a mobile hotspot on the same device. A hotspot changes the access medium, carrier, and network exit at once. That can show a link to the access path, but it can't prove that a specific carrier's backbone is at fault.
- Ethernet is fast, but Wi-Fi on the same connection is still slow: Check the wireless signal and local network load first. Pause uploads on other devices, move closer to the access point, and retest. A saturated upstream connection can also cause download stalls as acknowledgments get queued.
- Both nodes are slow on the original network but recover on another: Test a separate HTTPS target that you can access directly. If direct access is slow too, troubleshoot the access network first. If it works normally, keep the proxy-path comparison results; one direct speed test can't rule out every network-path issue.
- A traceroute shows packet loss at one hop: Intermediate devices may limit probe responses. If later hops and the destination don't show sustained loss, don't assume that hop is the fault. Focus on the destination's responses and whether actual downloads worsen at the same time.
Layer 3: Check routing and DNS before MTU
This flow describes a request entering the client. The system proxy affects only apps that use that proxy setting; TUN can take over more traffic at the system routing layer. DNS resolution may happen in the app, the operating system, or the core, so it doesn't necessarily occur at one fixed point before routing.
- Check routing: At the bottom of the main window, confirm which Routing profile is selected. Reproduce the request and, if logs are available, check the target and the outbound tag actually used. With rules evaluated in order, a broad rule can match before a more specific one. If the speed-test domain is routed directly, fix the rule before comparing node speeds.
- Check DNS: If pages are mainly slow to open, find out whether the delay occurs during resolution, connection, or the server response. Check whether the app uses its own secure DNS and whether the core resolves the target. Change only one setting that actually handles resolution per round, then retest with a new connection. Changing resolvers can also send the same site to a different content delivery node, so higher throughput doesn't necessarily mean faster DNS lookups.
- Check MTU last: Investigate the virtual adapter's MTU only if small requests work, large downloads repeatedly stall, and downloads from an app explicitly using the proxy work normally with TUN off. If your version has an MTU setting, note its original value. If it's 1500, try 1400 temporarily, restart TUN, and repeat the same download. Restore the original value if the improvement isn't repeatable. 1400 is a troubleshooting comparison value, not a universal recommendation.
Log errors and how to address them
First separate failed requests from slow ones. The errors below indicate connection or listening failures. Match log entries to the time the request was made so background updates or other apps' errors aren't mistaken for this speed test.
Error: failed to find an available destination
Cause: This error means an outbound attempt couldn't find an available destination. Check the underlying error in the same log entry for the specific cause. Fix: If the underlying error is no such host, check the server address and DNS results. If it times out, keep the node fixed and retest on another access network.
Error: bind: address already in use
Cause: The listening address and port are already in use, so the current core may not have started its inbound listener. Fix: Check the local port under Settings → Parameter settings, close any duplicate instances, or switch to an unused port and update any apps where you entered that port manually.
Common speed questions
The node shows 40 ms. Why are downloads still slow?
Compare two nodes using the same HTTPS file, and record time to first byte and sustained speed for each. A 40 ms reading describes only that probe. If the first byte arrives quickly but the download stays slow, check the exit path, network route, and target-side throttling before changing DNS.
The browser is fast, but my download tool is slow. What should I check first?
First check whether the download tool uses the system proxy or has the correct proxy type and local port configured manually. Turn off multi-connection downloads and compare the same file. If the two apps use different outbounds, make them use the same one before testing.
Several changes made things faster. How can I tell which setting helped?
Return to your recorded baseline, then apply one change at a time. Keep the target, node, and access network fixed, and repeat the test. Keep a change only if restoring the original value makes speeds drop and applying the change again brings them back. Revert any other changes that haven't been validated by retesting.