Updated July 2026 · 6 min read
Understand latency tests
The millisecond figure next to a node answers one question: how long a round trip to the test target takes through that node. It is not bandwidth, and it is not how fast a particular site will load for you.
What the number actually measures
The client sends a request to the test address, routes it through the node and back, and records how long the trip took. It reflects handshake and forwarding responsiveness — roughly, how quickly a page reacts the instant you click a link.
Bandwidth is nowhere in that measurement. A node at 80 ms may only manage two or three megabits, while a 300 ms veteran line saturates your download. Whether video stutters is something you only learn by pulling a large file through it yourself.
Change the test URL, change the results
The test address is configurable. Different targets sit with different providers in different data centres, one in the United States, another behind a global CDN, so the same node produces a different round trip against each of them.
That is why a screenshot posted by someone else proves nothing about your setup, and why a morning run against one URL cannot be lined up next to an afternoon run against another. Fix one address and test the whole batch through it.
Numbers are only comparable when the same batch of nodes is tested at the same moment against the same test URL. Two sets measured ten minutes apart tell you nothing.
A timeout does not mean a dead node
When a node shows a timeout or a blank cell, the first instinct is to write it off. More often the test address itself is unreachable from where you are, or the exit server blocks that particular target, while the node keeps working fine.
The check is simple: retest against a different address, or just switch to the node and open a web page. If pages load, the node is usable and the empty cell does not matter. The reverse case, a pretty number with nothing loading, is the one worth investigating.
Testing is real traffic too
Every test is a genuine connection. It shows up on the connections page and it counts against your quota. Firing a full subscription of several hundred nodes at once creates several hundred concurrent connections, and on a strained link they slow each other down.
For accurate figures, test in batches, or only test the groups you actually use. Repeated full sweeps also produce odd connection spikes on the provider side, and there is no everyday reason to press that button a dozen times a day.
What is the project?
Clash Verge Rev is a Tauri-based GUI for Mihomo with profile management, system proxy, TUN mode, rule editing and WebDAV backups.