Reading the Route Spectrum From Overseas Servers to Mainland China: CN2, CMI, 9929, 4837, and Japanese Upstreams

Reading the Route Spectrum From Overseas Servers to Mainland China: CN2, CMI, 9929, 4837, and Japanese Upstreams

The question "is this route good" is hard to answer because a route is not one object; it is a relay between autonomous systems. On the same Tokyo machine, Telecom traffic may enter China through NTT, Mobile traffic may ride CMI into the Mobile network, and Unicom traffic may loop through the United States on the way back. Each path has its own bottleneck, and any single sentence summarising "fast or slow" is guaranteed to be misleading.

The right order is therefore: establish which carrier your users are on, then look at this machine's actual path in that direction, and only then compare latency and throughput. Skipping the first two steps and reading a ping number yields nothing useful, because one machine can differ by more than a factor of two across three directions.

This guide breaks the China-bound path into an identifiable spectrum: the three-layer structure of any cross-border route, the signature networks of Telecom, Unicom, and Mobile, the differences among Japanese and Asia-Pacific upstreams, and a test-and-read method you can run. The GT-versus-GIA tiers inside the CN2 family are not covered here.

1. Break the Route Into Three Layers

1.1 Local Access, International Egress, and the Return Leg

Any cross-border path splits into three segments: access from the facility to the local exchange, international egress from the overseas upstream to a cable landing, and the backbone segment once inside mainland China. Congestion in any one of them degrades the whole experience, and sales copy tends to emphasise exactly one — usually the return leg. That is the most common reason people ask why an "optimized route" still feels slow.

Once split, troubleshooting gains an order: check whether local access detours (two facilities in one city can differ sharply), then which upstream carries the international egress, then which backbone the return leg lands on. Only the three together make a complete path.

1.2 Why One Machine Behaves Differently per Carrier

Carriers differ in where they peer abroad and how much capacity they hold there. Telecom concentrates its international egress at a few landing points, so its users abroad tend to favour upstreams with direct Telecom connectivity; Mobile built its international network earlier and routes with visibly different policies; Unicom depends on third-party transit in some directions. The result is a common pattern: normal latency toward Telecom, acceptable toward Mobile, and a detour through a third country toward Unicom, all on the same machine.

This also explains why "three-network optimization" is a promise with a cost. It does not happen automatically on a single machine; it happens when the facility buys capacity and arranges paths for each carrier separately. Understanding that makes the slogan of "full optimization" much less persuasive. For Hong Kong-specific choices, see how to read Hong Kong server routes.

2. The Telecom Spectrum

2.1 How the 163 Backbone Relates to CN2

The 163 backbone (AS4134) is Telecom's main carrier of public internet traffic: broad coverage, large capacity, but unstable under congestion. CN2 (AS4809) is a parallel premium backbone: smaller, higher priority, and more expensive. Without a CN2 product, Telecom users ride 163, and seeing the 202.97 range in a trace is normal rather than a fault.

CN2 itself comes in product tiers whose behaviour under congestion differs clearly; the details and verification methods are in the CN2 GIA line guide. For this article one fact suffices: CN2 is a paid priority lane inside Telecom's network, not a separate internet.

2.2 Reading Telecom-Specific Claims

Phrases like "Telecom optimized", "direct CN2", or "three-network optimization including Telecom" mean nothing until they land on an actual path. The method is to trace from a Telecom exit and see whether the path enters the signature CN2 range soon after leaving the country, and whether redundant hops through Japan, Korea, or Hong Kong appear.

Any "optimization" without a path description should be treated as unoptimized. That stance saves the most time, because paths are facts while adjectives are not.

3. The Unicom Spectrum

3.1 Where AS9929 and AS4837 Sit

AS9929 is commonly described as Unicom's premium international route: a comparatively clean path that holds up better at evening peak than ordinary AS4837. AS4837 is Unicom's public backbone, large in capacity but lower in priority, and more prone to queueing under congestion. Few overseas facilities can reliably supply AS9929 capacity, which is why such products command an obvious premium; AS4837 is the default Unicom path at most facilities.

Note that AS9929 is also heavily abused as a marketing word. Whether a path really uses 9929 can only be judged from the autonomous system numbers in a trace, never from a product title. For observations on how 9929 actually behaves, see notes on 9929 route stability.

3.2 Typical Unicom Detours

In some overseas regions Unicom lacks direct capacity, so traffic may return through third-party transit. The symptom is a hop in a country unrelated to the destination and a visible latency step — for instance a Los Angeles machine whose Unicom path goes to Japan before entering China, adding a hundred-odd milliseconds over the Telecom direction.

Unicom detours also shift with scheduling, so tests must cover evening peak and weekends. A clean path measured one day is not a long-term commitment.

4. The Mobile Spectrum

4.1 AS58453 CMI and CMIN2

AS58453 is China Mobile International's network. Mobile has expanded its self-built cross-border capacity quickly in recent years, and in some directions it now matches or beats Telecom and Unicom, which turned CMI-class products from niche to popular. CMIN2 is a further optimized tier on top of CMI, aimed mainly at evening-peak stability and consistent paths.

To check whether traffic really uses CMI, look for the address ranges CMI commonly uses and whether the path enters the mainland directly. For how the Mobile direction feels in practice, see the observation angles in hands-on notes on CMIN2 routes.

4.2 Direct Mobile Peering Versus Ordinary Transit

Direct Mobile peering means the facility interconnects with Mobile's international network itself: a short and predictable path. Ordinary transit may pass a third party before entering the Mobile network, making the path longer and dependent on that party's capacity. Idle-hour latency may look similar, but congested hours separate them clearly — which is why one facility sells both "with Mobile optimization" and "without".

Evaluate the Mobile direction separately. If Mobile users make up more than roughly a third of your audience, the price difference for that tier is usually worth paying.

5. Japanese and Asia-Pacific Upstreams

5.1 NTT (AS2914)

NTT is one of Japan's largest international bandwidth providers: broad coverage, many peering points, present in nearly every major facility. Its strength is predictability, global reach, and extensive peering with other tier-one carriers; its weakness is that default routing is not necessarily optimal toward China, so the facility must engineer policy.

"Connects to NTT" is therefore a neutral fact, not a quality promise: it says the facility has serious international interconnection, while the actual path to China still depends on specific policy and capacity allocation.

5.2 IIJ (AS2497), SoftBank (AS17676), and BBIX

IIJ (AS2497) is a Japanese commercial network provider with a solid reputation and careful interconnection toward Asia; SoftBank (AS17676) holds strong capacity both domestically and internationally and is often regarded as a better upstream for Japan; BBIX is a Japanese internet exchange, and a facility peering at BBIX can interconnect directly with many networks and pick paths more flexibly.

Judging a Japanese machine by its upstreams is more useful than judging it by the facility brand. Two facilities in the same building with different upstreams can behave completely differently toward China.

5.3 How Upstream Choice Shapes the China Path

Take two Tokyo facilities, one on NTT and one on SoftBank: the Telecom user's actual path may be entirely different, one transiting the United States and the other entering China over a direct submarine cable. The divergence toward Mobile and Unicom can be larger still. That is why "Japanese facility" cannot be used as a single label.

For a broader assessment of Japan nodes, combine the framework in the Japan cloud server review and put route, latency, and cost on one table rather than comparing price alone.

6. Testing and Reading the Results

6.1 What ping, mtr, and traceroute Each Tell You

ping gives latency and loss statistics, traceroute gives the path, and mtr combines both, accumulating per-hop loss while it runs — the most practical tool for judging a route. The commands are simple:

ping -c 200 SERVER_IP
mtr -rwzc 100 SERVER_IP

Sampling is what matters. Run each destination from Telecom, Unicom, and Mobile exits, ideally using three test machines on different carriers or asking colleagues in different regions. Testing from your own single broadband connection only describes that connection. To understand how commercial network platforms structure these paths, the network documentation from providers such as Zenlayer is a useful reference.

6.2 Sampling by Carrier and Time Window

Cross-border quality varies by the hour. Fix three windows — morning, afternoon, and evening peak — observe for a week, and record mean latency, jitter, and loss for each set. Evening-peak data deserves more weight because it determines what most users actually experience.

Weekends can differ from weekdays, especially on routes carrying entertainment traffic. For a long-term judgement, cover at least one full weekend.

6.3 How Latency, Jitter, and Loss Affect Transfer

Latency sets the number of round trips; jitter and loss decide whether each round trip goes smoothly. Transport protocols are acutely sensitive to loss: one percent loss can cut single-connection throughput by more than half. A route with low latency but high jitter therefore performs clearly worse than a slightly slower but stable one in large file transfers, video calls, and database replication.

This is the most frequently overlooked criterion in route selection. Web workloads notice it less because each response is small — but as soon as sustained transfer is involved, stability outweighs latency.

7. Misconceptions and Conclusions

7.1 Low Latency Is Not Stable Latency

Latency figures on a product page usually come from the best idle-hour sample, while real experience is set by the worst window. When choosing a route, rank evening-peak jitter first, average latency second, and peak loss third. Reverse that order and the choice goes wrong.

7.2 Why a Single Test Means Nothing

One ping describes one minute of network state. Cross-border routing shifts with carrier scheduling, cable maintenance, and transient congestion. No conclusion drawn from a single test can support a purchasing decision, let alone a long-term commitment.

The reverse is also true: one bad test does not prove a route is broken forever. Distinguish occasional jitter, which may recover on its own, from structural detouring, which only a facility change can fix.

7.3 The Order of Priorities When Choosing a Route

It comes down to one sequence: determine which carrier your users are on, inspect the autonomous system path in that direction, compare jitter and loss at evening peak, and only then look at price and specs. Doing it backwards — price first, latency number second — is the most common mistake.

One practical addition: do not expect a single machine to serve all three carriers well. If your audience is mixed, dual routes or CDN splitting are usually more reliable and cheaper than chasing one "fully optimized" line. For evaluating providers, see the VPS provider evaluation framework, and for more related coverage browse the cloud servers category.

Source: 16IDC original, pricing, routing and terms are subject to each provider's official documentation and contract.