Why the same IP range can feel so different
Two overseas datacenters can behave completely differently from a Chinese carrier. The difference is the ASN (autonomous system number) your traffic passes through — that is what "route analysis" reads.
Common lines at a glance
- China Telecom: AS4134 (163 backbone, standard) / AS4809 (CN2) / AS23764 (CN2 GIA, premium international);
- China Unicom: AS4837 (169 backbone) / AS9929 (CUII, premium);
- China Mobile: AS9808 (CMNET) / AS58453 (CMI, international egress).
How to measure
- Take the target IP your datacenter assigned;
- Run traceroute / mtr from Telecom, Unicom and Mobile probes;
- Read the ASN of each hop: 4809 means CN2, 9929 means CUII, and only 4134 means the standard 163 backbone.
Reading the result
The outbound path is decided by the Chinese carrier egress; the return path by the datacenter upstream — they can differ. 100% loss on intermediate hops usually means those devices drop ICMP replies, not that the path is broken; judge by the last hops plus overall latency and loss.
How the lines compare
| Line | ASN | Peak-hour behaviour | Price tier |
|---|---|---|---|
| Telecom 163 backbone | AS4134 | Congestion and visible packet loss | Low |
| CN2 GT | AS4809 | Acceptable; the return path may drop to 163 | Medium |
| CN2 GIA | AS4809 / AS23764 | Stable and low latency both ways | High |
| Unicom 9929 (CUII) | AS9929 | Excellent from Unicom | Medium-high |
| Mobile CMI | AS58453 | Friendly to Mobile users | Medium |
Verifying a vendor's claim
- Ask for a test IP — a demo page is often rate-limited or specially optimized;
- Run mtr from Telecom, Unicom and Mobile probes and check that both directions use the premium route;
- Watch latency and loss between 20:00 and 23:00; daytime speed numbers prove little;
- Routing changes, so one test reflects only that moment — sample a few days.
Follow-up questions
Why does the return path matter more? Perceived download and page-load speed is driven by the return path (data coming back to you). Premium outbound with a 163 return still feels poor. Does 100% loss on an intermediate hop matter? Usually not — backbone devices commonly ignore ICMP; judge by the last hops and overall loss.
Three tips when buying
- Get a test IP before paying: watch actual behaviour across several peak evenings instead of the vendor's speed-test screenshots;
- Judge each carrier separately: the same datacenter can behave very differently for Telecom, Unicom and Mobile — choose according to your user mix;
- Focus on the return path: downloads and page loads are shaped by the return path; premium outbound with an ordinary return still feels mediocre.
Try it: IP lookup (with route analysis)
Reading route quality metrics
- Latency: look at mean and jitter (P95 minus P50); reporting only the mean hides peak-hour degradation;
- Loss: sustained loss above 0.5% noticeably cuts TCP throughput because it collapses the congestion window;
- Hops and detours: fewer hops is usually steadier at equal latency, but check for congested intermediate nodes;
- Sample across the day: evening peaks and small hours can differ widely — one run proves little; compare several windows.