All Vultr plans

Where the test IPs and latency figures come from

A test IP is a machine the vendor publishes for this data centre and this route. Before buying you can ping it to see how fast your own network reaches it. Clicking an IP opens a multi-node ping across China on itdog, which you can break down by province and carrier.

Where the IPs come from

The vendor's website or Looking Glass page, or the vendor's reply to a support ticket; a few come from third-party reviews. When an IP published by the vendor no longer answers, we replace it with the reachable address with the lowest latency in the same subnet.

Who measures the latency

Probes in mainland China from two public measurement networks. Each hour one vendor takes its turn, so each vendor is measured in two or three time slots a day, by day and at night, and the day's runs are averaged. Each run uses up to 12 probes on each of Telecom, Unicom and Mobile (6 per network; when a network has fewer that day, we use what it has), and each probe pings 4 times.

How to read it

“Telecom 18 · Unicom 22 · Mobile 37 ms” is the latency on each of the three carriers; hover to see the probe cities and loss rate. When a carrier shows 20% loss or more, it is spelt out, e.g. “Telecom 99 (loss 55%)”. That latency figure is only the average of the packets that came back, and in real use the connection will lag; the “My network” filter and sorting do not count it as reachable. We have cross-checked such loss over TCP ports: the handshake time grows several-fold, so it is the path from that carrier to the data centre that drops packets, not the server rate-limiting ping. Sorting by the column uses the fastest of the three carriers.

No reply and unreachable

“no reply” on one carrier means that probe could not reach the IP. “Test IP unreachable” means probes both inside and outside China could not reach it, usually because the IP has been retired; it does not mean the route or the server is unavailable. IPs that block ping are tested on TCP port 80/443 instead.

How accurate is it

For reference only. Each run uses at most 12 probes per carrier, pinging 4 times each; the probes change every run, most sit in data centres rather than on home broadband, and bandwidth and the evening peak are not measured.

What the route names mean

The route is which carrier's network, and which of its networks, carries traffic between mainland China and the data centre. The closer to China and the better the route, the smoother SSH and web pages feel, and the steadier things are at the evening peak.

CN2 GIA

China Telecom's premium network (AS4809), with CN2 both ways. Low latency, stable at the evening peak, the most expensive. The first choice for Telecom users.

CN2 GT

The standard version of Telecom CN2: the international leg uses CN2, the domestic leg uses the ordinary 163 backbone. Cheaper than GIA, a little less stable at the evening peak.

CN2

The vendor only says CN2, without stating GIA or GT; it is safer to expect GT-level performance.

9929

China Unicom's premium network (AS9929, also called Unicom A network or CUII). Works best for Unicom users.

4837

China Unicom's standard backbone (AS4837). Cheap, plentiful bandwidth; may be congested at the evening peak.

CMIN2

China Mobile's premium network (AS58807). Works best for Mobile users.

CMI

China Mobile International's standard outbound route (AS58453): better than a standard route for Mobile users, worse than CMIN2.

All 3 carriers optimised

The vendor states that return traffic for all of Telecom, Unicom and Mobile goes over optimised routes, so it will not be too bad whichever broadband you use.

Standard route

A standard international route, not optimised for mainland China; may slow down noticeably at the evening peak. Suitable as a Main server, reached through a Relay.

This site labels routes as each vendor states them and does not run route tests. Real-world results depend on your broadband carrier; test with the test IP before buying.