VPS Speed Test Nodes & Carrier Route Benchmarks

For traffic between mainland China and an overseas VPS, what decides the experience is not the data centre's location but the return route — the path packets take back to your broadband. Two nodes in the same Singapore facility can differ by more than 100 ms at peak depending on whether they return via CN2 GIA or generic BGP.

How to read the results above

Latency (RTT) reflects physical distance and how much the path detours; lines in the same facility can differ 3–5×. Jitter is the variance in latency — above 20 ms it noticeably hurts voice and gaming. Packet loss usually matters more than latency: sustained loss above 1% can substantially reduce TCP throughput, especially on high-latency paths, and rising loss at peak usually indicates a congested egress. Download speed is capped by your own access line, so compare nodes against each other rather than reading absolute numbers.

Latency: 6 uncacheable 256 B probes per candidate (timeout 2500 ms); sample 0 is discarded because it pays the TCP/TLS handshake; the median of the rest is reported. Jitter: mean absolute difference between successive sorted samples. Probe failure: a probe that times out or throws, shown as "failed / total"; browsers cannot send ICMP, so this is not ping-level packet loss. Promotion: once node-direct test hostnames are live, the 5 lowest measured-latency candidates advance to a 1 MB per-node throughput tier served by the node itself; today no node-direct hostname exists, so the only throughput measurement is your access-line baseline (10 MB via the Cloudflare edge). The 100 MB tier always requires an explicit confirmation. Throughput: the first 256 KB of slow-start bytes are discarded, then bytes/time over a window of at most 8 s. Every latency probe carries a unique cache-buster and no-store, so probe timing never hits a cache. Throughput reference files are deliberately served from Cloudflare edge cache: a download measures your access path to the edge (the same path the probes measure) and keeps per-test Worker egress out of the timing. Cloudflare currently provides the control plane (catalogue, ASN/PoP hints, ranking) and the access-line baseline measurement (probes and edge-cached reference files at your nearest PoP); until node-direct test hostnames are provisioned, every browser figure describes your path to the Cloudflare edge, labelled "via Cloudflare edge", while per-node rows are labelled "predicted". Node-specific direct measurement becomes available once those hostnames are provisioned. Browser limitations: no ICMP, no control over socket concurrency, and your own access line caps every throughput figure.