Which VPS datacenter is closest to your readers?

How we measure

• Targets. Each provider’s own public speed-test or looking-glass host for that region (Vultr, Akamai/Linode, OVHcloud, Scaleway, Hetzner). DigitalOcean no longer runs public speed-test hosts, so we use its Spaces endpoint in the same datacenter; AWS Lightsail, Alibaba Cloud and Tencent Cloud use their official regional endpoints. Every host and its source page is listed below.
• Probes. Globalping community probes, picked by city or by carrier (China Telecom AS4134, China Unicom AS4837, China Mobile AS9808). The same probes are reused for every target within a round when possible.
• Method. 10 pings per pair, twice a day (ICMP; TCP port 443 for Lightsail, which drops ICMP). We store the median round-trip time, packet loss and jitter (mean change between consecutive pings). The matrix shows the median of the last 14 rounds (7 days). Once a week we run mtr from the China probes to the Asian and US-West targets to label the route.
• Budget. The free Globalping API allows 250 probe-tests an hour. A full round costs about 901 tests, so it is spread over a few hours; 1802 tests were used in the last 24 hours.

Limitations

• Most China probes sit in data centres (5 of 8 columns in the last round). Home broadband, especially in the evening, is usually slower and loses more packets. Use the numbers to rank datacenters, not as a promise of what your readers will see.
• One probe per cell. A single probe can have a bad day. Look at the trend and the loss column before you decide.
• A test host is not your VM. Your server may sit on a different network segment, and premium China routes (CN2 GIA, CMI, 9929) are often sold as separate products. Route labels come from AS numbers in the traceroute and are only a hint.
• Ping is not page speed. Throughput, TLS, your CDN and the Great Firewall all matter. For readers in China, a CDN or a Hong Kong/Japan server with a premium route often beats raw distance.